Authors: Nan Li, Xiaolu Wang, Xiao Du, Puyu Cai, Ting Wang
Abstract: Federated Learning (FL) enables clients to collaboratively train machine learning models without sharing local data, preserving privacy in diverse environments. While traditional FL approaches preserve privacy, they often struggle with high computational and communication overhead. To address these issues, model pruning is introduced as a strategy to streamline computations. However, existing pruning methods, when applied solely based on local data, often produce sub-models that inadequately reflect clients' specific tasks due to data insufficiency. To overcome these challenges, this paper introduces SAFL (Structure-Aware Federated Learning), a novel framework that enhances personalized federated learning through client-specific clustering and Similar Client Structure Information (SCSI)-guided model pruning. SAFL employs a two-stage process: initially, it groups clients based on data similarities and uses aggregated pruning criteria to guide the pruning process, facilitating the identification of optimal sub-models. Subsequently, clients train these pruned models and engage in server-based aggregation, ensuring tailored and efficient models for each client. This method significantly reduces computational overhead while improving inference accuracy. Extensive experiments demonstrate that SAFL markedly diminishes model size and improves performance, making it highly effective in federated environments characterized by heterogeneous data.
Authors: Dario Coscia, Max Welling, Nicola Demo, Gianluigi Rozza
Abstract: Autoregressive and recurrent networks have achieved remarkable progress across various fields, from weather forecasting to molecular generation and Large Language Models. Despite their strong predictive capabilities, these models lack a rigorous framework for addressing uncertainty, which is key in scientific applications such as PDE solving, molecular generation and Machine Learning Force Fields. To address this shortcoming we present BARNN: a variational Bayesian Autoregressive and Recurrent Neural Network. BARNNs aim to provide a principled way to turn any autoregressive or recurrent model into its Bayesian version. BARNN is based on the variational dropout method, allowing to apply it to large recurrent neural networks as well. We also introduce a temporal version of the "Variational Mixtures of Posteriors" prior (tVAMP-prior) to make Bayesian inference efficient and well-calibrated. Extensive experiments on PDE modelling and molecular generation demonstrate that BARNN not only achieves comparable or superior accuracy compared to existing methods, but also excels in uncertainty quantification and modelling long-range dependencies.
Authors: Einar Urdshals, Jasmina Urdshals
Abstract: We study how a one-layer attention-only transformer develops relevant structures while learning to sort lists of numbers. At the end of training, the model organizes its attention heads in two main modes that we refer to as vocabulary-splitting and copy-suppression. Both represent simpler modes than having multiple heads handle overlapping ranges of numbers. Interestingly, vocabulary-splitting is present regardless of whether we use weight decay, a common regularization technique thought to drive simplification, supporting the thesis that neural networks naturally prefer simpler solutions. We relate copy-suppression to a mechanism in GPT-2 and investigate its functional role in our model. Guided by insights from a developmental analysis of the model, we identify features in the training data that drive the model's final acquired solution. This provides a concrete example of how the training data shape the internal organization of transformers, paving the way for future studies that could help us better understand how LLMs develop their internal structures.
Authors: Jesus Paucar-Escalante, Matheus Alves da Silva, Bruno De Lima Sanches, Aurea Soriano-Vargas, Laura Silveira Moriyama, Esther Luna Colombini
Abstract: Parkinson's disease (PD) is a neurological disorder requiring early and accurate diagnosis for effective management. Machine learning (ML) has emerged as a powerful tool to enhance PD classification and diagnostic accuracy, particularly by leveraging wearable sensor data. This survey comprehensively reviews current ML methodologies used in classifying Parkinsonian tremors, evaluating various tremor data acquisition methodologies, signal preprocessing techniques, and feature selection methods across time and frequency domains, highlighting practical approaches for tremor classification. The survey explores ML models utilized in existing studies, ranging from traditional methods such as Support Vector Machines (SVM) and Random Forests to advanced deep learning architectures like Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM). We assess the efficacy of these models in classifying tremor patterns associated with PD, considering their strengths and limitations. Furthermore, we discuss challenges and discrepancies in current research and broader challenges in applying ML to PD diagnosis using wearable sensor data. We also outline future research directions to advance ML applications in PD diagnostics, providing insights for researchers and practitioners.
Authors: Matan Ben-Dov, Jing Chen
Abstract: Tensor-network Born machines (TNBMs) are quantum-inspired generative models for learning data distributions. Using tensor-network contraction and optimization techniques, the model learns an efficient representation of the target distribution, capable of capturing complex correlations with a compact parameterization. Despite their promise, the optimization of TNBMs presents several challenges. A key bottleneck of TNBMs is the logarithmic nature of the loss function that is commonly used for this problem. The single-tensor logarithmic optimization problem cannot be solved analytically, necessitating an iterative approach that slows down convergence and increases the risk of getting trapped in one of many non-optimal local minima. In this paper, we present an improved second-order optimization technique for TNBM training, which significantly enhances convergence rates and the quality of the optimized model. Our method employs a modified Newton's method on the manifold of normalized states, incorporating regularization of the loss landscape to mitigate local minima issues. We demonstrate the effectiveness of our approach by training a one-dimensional matrix product state (MPS) on both discrete and continuous datasets, showcasing its advantages in terms of stability, efficiency, and generalization.
Authors: Hugo Inzirillo, Remi Genet
Abstract: Traditional Smooth Transition Autoregressive (STAR) models offer an effective way to model these dynamics through smooth regime changes based on specific transition variables. In this paper, we propose a novel approach by drawing an analogy between STAR models and a multilayer neural network architecture. Our proposed neural network architecture mimics the STAR framework, employing multiple layers to simulate the smooth transition between regimes and capturing complex, nonlinear relationships. The network's hidden layers and activation functions are structured to replicate the gradual switching behavior typical of STAR models, allowing for a more flexible and scalable approach to regime-dependent modeling. This research suggests that neural networks can provide a powerful alternative to STAR models, with the potential to enhance predictive accuracy in economic and financial forecasting.
Authors: Devansh Bhardwaj, Naman Mishra
Abstract: Fingerprinting refers to the process of identifying underlying Machine Learning (ML) models of AI Systemts, such as Large Language Models (LLMs), by analyzing their unique characteristics or patterns, much like a human fingerprint. The fingerprinting of Large Language Models (LLMs) has become essential for ensuring the security and transparency of AI-integrated applications. While existing methods primarily rely on access to direct interactions with the application to infer model identity, they often fail in real-world scenarios involving multi-agent systems, frequent model updates, and restricted access to model internals. In this paper, we introduce a novel fingerprinting framework designed to address these challenges by integrating static and dynamic fingerprinting techniques. Our approach identifies architectural features and behavioral traits, enabling accurate and robust fingerprinting of LLMs in dynamic environments. We also highlight new threat scenarios where traditional fingerprinting methods are ineffective, bridging the gap between theoretical techniques and practical application. To validate our framework, we present an extensive evaluation setup that simulates real-world conditions and demonstrate the effectiveness of our methods in identifying and monitoring LLMs in Gen-AI applications. Our results highlight the framework's adaptability to diverse and evolving deployment contexts.
Authors: Harshwardhan Praveen, Jacob Brown, Christopher Earls
Abstract: In this work, we present a mesh-independent, data-driven library, chebgreen, to mathematically model one-dimensional systems, possessing an associated control parameter, and whose governing partial differential equation is unknown. The proposed method learns an Empirical Green's Function for the associated, but hidden, boundary value problem, in the form of a Rational Neural Network from which we subsequently construct a bivariate representation in a Chebyshev basis. We uncover the Green's function, at an unseen control parameter value, by interpolating the left and right singular functions within a suitable library, expressed as points on a manifold of Quasimatrices, while the associated singular values are interpolated with Lagrange polynomials.
Authors: Maria R. Lima, Alexander Capstick, Fatemeh Geranmayeh, Ramin Nilforooshan, Maja Matari\'c, Ravi Vaidyanathan, Payam Barnaghi
Abstract: Timely and accurate assessment of cognitive impairment is a major unmet need in populations at risk. Alterations in speech and language can be early predictors of Alzheimer's disease and related dementias (ADRD) before clinical signs of neurodegeneration. Voice biomarkers offer a scalable and non-invasive solution for automated screening. However, the clinical applicability of machine learning (ML) remains limited by challenges in generalisability, interpretability, and access to patient data to train clinically applicable predictive models. Using DementiaBank recordings (N=291, 64% female), we evaluated ML techniques for ADRD screening and severity prediction from spoken language. We validated model generalisability with pilot data collected in-residence from older adults (N=22, 59% female). Risk stratification and linguistic feature importance analysis enhanced the interpretability and clinical utility of predictions. For ADRD classification, a Random Forest applied to lexical features achieved a mean sensitivity of 69.4% (95% confidence interval (CI) = 66.4-72.5) and specificity of 83.3% (78.0-88.7). On real-world pilot data, this model achieved a mean sensitivity of 70.0% (58.0-82.0) and specificity of 52.5% (39.3-65.7). For severity prediction using Mini-Mental State Examination (MMSE) scores, a Random Forest Regressor achieved a mean absolute MMSE error of 3.7 (3.7-3.8), with comparable performance of 3.3 (3.1-3.5) on pilot data. Linguistic features associated with higher ADRD risk included increased use of pronouns and adverbs, greater disfluency, reduced analytical thinking, lower lexical diversity and fewer words reflecting a psychological state of completion. Our interpretable predictive modelling offers a novel approach for in-home integration with conversational AI to monitor cognitive health and triage higher-risk individuals, enabling earlier detection and intervention.
Authors: Zehong Wang, Zheyuan Zhang, Tianyi Ma, Nitesh V Chawla, Chuxu Zhang, Yanfang Ye
Abstract: Graph learning tasks require models to comprehend essential substructure patterns relevant to downstream tasks, such as triadic closures in social networks and benzene rings in molecular graphs. Due to the non-Euclidean nature of graphs, existing graph neural networks (GNNs) rely on message passing to iteratively aggregate information from local neighborhoods. Despite their empirical success, message passing struggles to identify fundamental substructures, such as triangles, limiting its expressiveness. To overcome this limitation, we propose the Neural Graph Pattern Machine (GPM), a framework designed to learn directly from graph patterns. GPM efficiently extracts and encodes substructures while identifying the most relevant ones for downstream tasks. We also demonstrate that GPM offers superior expressivity and improved long-range information modeling compared to message passing. Empirical evaluations on node classification, link prediction, graph classification, and regression show the superiority of GPM over state-of-the-art baselines. Further analysis reveals its desirable out-of-distribution robustness, scalability, and interpretability. We consider GPM to be a step toward going beyond message passing.
Authors: Dan Liu, Samer El Kababji, Nicholas Mitsakakis, Lisa Pilgram, Thomas Walters, Mark Clemons, Greg Pond, Alaa El-Hussuna, Khaled El Emam
Abstract: Small datasets are common in health research. However, the generalization performance of machine learning models is suboptimal when the training datasets are small. To address this, data augmentation is one solution. Augmentation increases sample size and is seen as a form of regularization that increases the diversity of small datasets, leading them to perform better on unseen data. We found that augmentation improves prognostic performance for datasets that: have fewer observations, with smaller baseline AUC, have higher cardinality categorical variables, and have more balanced outcome variables. No specific generative model consistently outperformed the others. We developed a decision support model that can be used to inform analysts if augmentation would be useful. For seven small application datasets, augmenting the existing data results in an increase in AUC between 4.31% (AUC from 0.71 to 0.75) and 43.23% (AUC from 0.51 to 0.73), with an average 15.55% relative improvement, demonstrating the nontrivial impact of augmentation on small datasets (p=0.0078). Augmentation AUC was higher than resampling only AUC (p=0.016). The diversity of augmented datasets was higher than the diversity of resampled datasets (p=0.046).
Authors: Zijun Deng, Rafael Orozco, Abhinav Prakash Gahlot, Felix J. Herrmann
Abstract: Reducing CO$_2$ emissions is crucial to mitigating climate change. Carbon Capture and Storage (CCS) is one of the few technologies capable of achieving net-negative CO$_2$ emissions. However, predicting fluid flow patterns in CCS remains challenging due to uncertainties in CO$_2$ plume dynamics and reservoir properties. Building on existing seismic imaging methods like the Joint Recovery Method (JRM), which lacks uncertainty quantification, we propose the Probabilistic Joint Recovery Method (pJRM). By estimating posterior distributions across surveys using a shared generative model, pJRM provides uncertainty information to improve risk assessment in CCS projects.
Authors: Michael S. Yao, James C. Gee, Osbert Bastani
Abstract: The goal of offline model-based optimization (MBO) is to propose new designs that maximize a reward function given only an offline dataset. However, an important desiderata is to also propose a diverse set of final candidates that capture many optimal and near-optimal design configurations. We propose Diversity in Adversarial Model-based Optimization (DynAMO) as a novel method to introduce design diversity as an explicit objective into any MBO problem. Our key insight is to formulate diversity as a distribution matching problem where the distribution of generated designs captures the inherent diversity contained within the offline dataset. Extensive experiments spanning multiple scientific domains show that DynAMO can be used with common optimization methods to significantly improve the diversity of proposed designs while still discovering high-quality candidates.
Authors: Mrityunjay Sharma, Sarabeshwar Balaji, Pinaki Saha, Ritesh Kumar
Abstract: We explore a suite of generative modelling techniques to efficiently navigate and explore the complex landscapes of odor and the broader chemical space. Unlike traditional approaches, we not only generate molecules but also predict the odor likeliness with ROC AUC score of 0.97 and assign probable odor labels. We correlate odor likeliness with physicochemical features of molecules using machine learning techniques and leverage SHAP (SHapley Additive exPlanations) to demonstrate the interpretability of the function. The whole process involves four key stages: molecule generation, stringent sanitization checks for molecular validity, fragrance likeliness screening and odor prediction of the generated molecules. By making our code and trained models publicly accessible, we aim to facilitate broader adoption of our research across applications in fragrance discovery and olfactory research.
Authors: Alessio Russo, Alberto Maria Metelli, Marcello Restelli
Abstract: We tackle average-reward infinite-horizon POMDPs with an unknown transition model but a known observation model, a setting that has been previously addressed in two limiting ways: (i) frequentist methods relying on suboptimal stochastic policies having a minimum probability of choosing each action, and (ii) Bayesian approaches employing the optimal policy class but requiring strong assumptions about the consistency of employed estimators. Our work removes these limitations by proving convenient estimation guarantees for the transition model and introducing an optimistic algorithm that leverages the optimal class of deterministic belief-based policies. We introduce modifications to existing estimation techniques providing theoretical guarantees separately for each estimated action transition matrix. Unlike existing estimation methods that are unable to use samples from different policies, we present a novel and simple estimator that overcomes this barrier. This new data-efficient technique, combined with the proposed \emph{Action-wise OAS-UCRL} algorithm and a tighter theoretical analysis, leads to the first approach enjoying a regret guarantee of order $\mathcal{O}(\sqrt{T \,\log T})$ when compared against the optimal policy, thus improving over state of the art techniques. Finally, theoretical results are validated through numerical simulations showing the efficacy of our method against baseline methods.
Authors: Hanyang Wang, Juergen Branke, Matthias Poloczek
Abstract: Many real-world black-box optimization problems have multiple conflicting objectives. Rather than attempting to approximate the entire set of Pareto-optimal solutions, interactive preference learning allows to focus the search on the most relevant subset. However, few previous studies have exploited the fact that utility functions are usually monotonic. In this paper, we address the Bayesian Optimization with Preference Exploration (BOPE) problem and propose using a neural network ensemble as a utility surrogate model. This approach naturally integrates monotonicity and supports pairwise comparison data. Our experiments demonstrate that the proposed method outperforms state-of-the-art approaches and exhibits robustness to noise in utility evaluations. An ablation study highlights the critical role of monotonicity in enhancing performance.
Authors: Kelvin Kan, Xingjian Li, Stanley Osher
Abstract: Transformers have achieved state-of-the-art performance in numerous tasks. In this paper, we propose a continuous-time formulation of transformers. Specifically, we consider a dynamical system whose governing equation is parametrized by transformer blocks. We leverage optimal transport theory to regularize the training problem, which enhances stability in training and improves generalization of the resulting model. Moreover, we demonstrate in theory that this regularization is necessary as it promotes uniqueness and regularity of solutions. Our model is flexible in that almost any existing transformer architectures can be adopted to construct the dynamical system with only slight modifications to the existing code. We perform extensive numerical experiments on tasks motivated by natural language processing, image classification, and point cloud classification. Our experimental results show that the proposed method improves the performance of its discrete counterpart and outperforms relevant comparing models.
Authors: Qiyao Liang, Daoyuan Qian, Liu Ziyin, Ila Fiete
Abstract: Composition-the ability to generate myriad variations from finite means-is believed to underlie powerful generalization. However, compositional generalization remains a key challenge for deep learning. A widely held assumption is that learning disentangled (factorized) representations naturally supports this kind of extrapolation. Yet, empirical results are mixed, with many generative models failing to recognize and compose factors to generate out-of-distribution (OOD) samples. In this work, we investigate a controlled 2D Gaussian "bump" generation task, demonstrating that standard generative architectures fail in OOD regions when training with partial data, even when supplied with fully disentangled $(x, y)$ coordinates, re-entangling them through subsequent layers. By examining the model's learned kernels and manifold geometry, we show that this failure reflects a "memorization" strategy for generation through the superposition of training data rather than by combining the true factorized features. We show that models forced-through architectural modifications with regularization or curated training data-to create disentangled representations in the full-dimensional representational (pixel) space can be highly data-efficient and effective at learning to compose in OOD regions. These findings underscore that bottlenecks with factorized/disentangled representations in an abstract representation are insufficient: the model must actively maintain or induce factorization directly in the representational space in order to achieve robust compositional generalization.
Authors: Yerin Kim, Alexander Benvenuti, Bo Chen, Mustafa Karabag, Abhishek Kulkarni, Nathaniel D. Bastian, Ufuk Topcu, Matthew Hale
Abstract: Autonomous systems are increasingly expected to operate in the presence of adversaries, though an adversary may infer sensitive information simply by observing a system, without even needing to interact with it. Therefore, in this work we present a deceptive decision-making framework that not only conceals sensitive information, but in fact actively misleads adversaries about it. We model autonomous systems as Markov decision processes, and we consider adversaries that attempt to infer their reward functions using inverse reinforcement learning. To counter such efforts, we present two regularization strategies for policy synthesis problems that actively deceive an adversary about a system's underlying rewards. The first form of deception is ``diversionary'', and it leads an adversary to draw any false conclusion about what the system's reward function is. The second form of deception is ``targeted'', and it leads an adversary to draw a specific false conclusion about what the system's reward function is. We then show how each form of deception can be implemented in policy optimization problems, and we analytically bound the loss in total accumulated reward that is induced by deception. Next, we evaluate these developments in a multi-agent sequential decision-making problem with one real agent and multiple decoys. We show that diversionary deception can cause the adversary to believe that the most important agent is the least important, while attaining a total accumulated reward that is $98.83\%$ of its optimal, non-deceptive value. Similarly, we show that targeted deception can make any decoy appear to be the most important agent, while still attaining a total accumulated reward that is $99.25\%$ of its optimal, non-deceptive value.
Authors: Taehyeun Kim, Tae-Geun Kim, Anouck Girard, Ilya Kolmanovsky
Abstract: In this paper, we develop a neural network-based approach for time-series prediction in unknown Hamiltonian dynamical systems. Our approach leverages a surrogate model and learns the system dynamics using generalized coordinates (positions) and their conjugate momenta while preserving a constant Hamiltonian. To further enhance long-term prediction accuracy, we introduce an Autoregressive Hamiltonian Neural Network, which incorporates autoregressive prediction errors into the training objective. Additionally, we employ Bayesian data assimilation to refine predictions in real-time using online measurement data. Numerical experiments on a spring-mass system and highly elliptic orbits under gravitational perturbations demonstrate the effectiveness of the proposed method, highlighting its potential for accurate and robust long-term predictions.
Authors: Adam Scherlis, Nora Belrose
Abstract: We present an algorithm for estimating the probability mass, under a Gaussian or uniform prior, of a region in neural network parameter space corresponding to a particular behavior, such as achieving test loss below some threshold. When the prior is uniform, this problem is equivalent to measuring the volume of a region. We show empirically and theoretically that existing algorithms for estimating volumes in parameter space underestimate the true volume by millions of orders of magnitude. We find that this error can be dramatically reduced, but not entirely eliminated, with an importance sampling method using gradient information that is already provided by popular optimizers. The negative logarithm of this probability can be interpreted as a measure of a network's information content, in accordance with minimum description length (MDL) principles and rate-distortion theory. As expected, this quantity increases during language model training. We also find that badly-generalizing behavioral regions are smaller, and therefore less likely to be sampled at random, demonstrating an inductive bias towards well-generalizing functions.
Authors: Mohammad Fatahi, Danial Sadrian Zadeh, Benyamin Ghojogh, Behzad Moshiri, Otman Basir
Abstract: Autonomous vehicles represent a revolutionary advancement driven by the integration of artificial intelligence within intelligent transportation systems. However, they remain vulnerable due to the absence of robust security mechanisms in the Controller Area Network (CAN) bus. In order to mitigate the security issue, many machine learning models and strategies have been proposed, which primarily focus on a subset of dominant patterns of anomalies and lack rigorous evaluation in terms of reliability and robustness. Therefore, to address the limitations of previous works and mitigate the security vulnerability in CAN bus, the current study develops a model based on the intrinsic nature of the problem to cover all dominant patterns of anomalies. To achieve this, a cascade feature-level fusion strategy optimized by a two-parameter genetic algorithm is proposed to combine temporal and spatial information. Subsequently, the model is evaluated using a paired t-test to ensure reliability and robustness. Finally, a comprehensive comparative analysis conducted on two widely used datasets advocates that the proposed model outperforms other models and achieves superior accuracy and F1-score, demonstrating the best performance among all models presented to date.
Authors: Gon\c{c}alo Paulo, Stepan Shabalin, Nora Belrose
Abstract: Sparse autoencoders (SAEs) extract human-interpretable features from deep neural networks by transforming their activations into a sparse, higher dimensional latent space, and then reconstructing the activations from these latents. Transcoders are similar to SAEs, but they are trained to reconstruct the output of a component of a deep network given its input. In this work, we compare the features found by transcoders and SAEs trained on the same model and data, finding that transcoder features are significantly more interpretable. We also propose _skip transcoders_, which add an affine skip connection to the transcoder architecture, and show that these achieve lower reconstruction loss with no effect on interpretability.
Authors: Fan Wang, Feiyu Jiang, Zifeng Zhao, Yi Yu
Abstract: Dynamic pricing strategies are crucial for firms to maximize revenue by adjusting prices based on market conditions and customer characteristics. However, designing optimal pricing strategies becomes challenging when historical data are limited, as is often the case when launching new products or entering new markets. One promising approach to overcome this limitation is to leverage information from related products or markets to inform the focal pricing decisions. In this paper, we explore transfer learning for nonparametric contextual dynamic pricing under a covariate shift model, where the marginal distributions of covariates differ between source and target domains while the reward functions remain the same. We propose a novel Transfer Learning for Dynamic Pricing (TLDP) algorithm that can effectively leverage pre-collected data from a source domain to enhance pricing decisions in the target domain. The regret upper bound of TLDP is established under a simple Lipschitz condition on the reward function. To establish the optimality of TLDP, we further derive a matching minimax lower bound, which includes the target-only scenario as a special case and is presented for the first time in the literature. Extensive numerical experiments validate our approach, demonstrating its superiority over existing methods and highlighting its practical utility in real-world applications.
Authors: Gon\c{c}alo Paulo, Nora Belrose
Abstract: The greatest ambition of mechanistic interpretability is to completely rewrite deep neural networks in a format that is more amenable to human understanding, while preserving their behavior and performance. In this paper, we attempt to partially rewrite a large language model using simple natural language explanations. We first approximate one of the feedforward networks in the LLM with a wider MLP with sparsely activating neurons - a transcoder - and use an automated interpretability pipeline to generate explanations for these neurons. We then replace the first layer of this sparse MLP with an LLM-based simulator, which predicts the activation of each neuron given its explanation and the surrounding context. Finally, we measure the degree to which these modifications distort the model's final output. With our pipeline, the model's increase in loss is statistically similar to entirely replacing the sparse MLP output with the zero vector. We employ the same protocol, this time using a sparse autoencoder, on the residual stream of the same layer and obtain similar results. These results suggest that more detailed explanations are needed to improve performance substantially above the zero ablation baseline.
Authors: Wojciech Zaremba, Evgenia Nitishinskaya, Boaz Barak, Stephanie Lin, Sam Toyer, Yaodong Yu, Rachel Dias, Eric Wallace, Kai Xiao, Johannes Heidecke, Amelia Glaese
Abstract: We conduct experiments on the impact of increasing inference-time compute in reasoning models (specifically OpenAI o1-preview and o1-mini) on their robustness to adversarial attacks. We find that across a variety of attacks, increased inference-time compute leads to improved robustness. In many cases (with important exceptions), the fraction of model samples where the attack succeeds tends to zero as the amount of test-time compute grows. We perform no adversarial training for the tasks we study, and we increase inference-time compute by simply allowing the models to spend more compute on reasoning, independently of the form of attack. Our results suggest that inference-time compute has the potential to improve adversarial robustness for Large Language Models. We also explore new attacks directed at reasoning models, as well as settings where inference-time compute does not improve reliability, and speculate on the reasons for these as well as ways to address them.
Authors: Han Zhong, Yutong Yin, Shenao Zhang, Xiaojun Xu, Yuanxin Liu, Yifei Zuo, Zhihan Liu, Boyi Liu, Sirui Zheng, Hongyi Guo, Liwei Wang, Mingyi Hong, Zhaoran Wang
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities in complex reasoning tasks, yet generating reliable reasoning processes remains a significant challenge. We present a unified probabilistic framework that formalizes LLM reasoning through a novel graphical model incorporating latent thinking processes and evaluation signals. Within this framework, we introduce the Bootstrapping Reinforced Thinking Process (BRiTE) algorithm, which works in two steps. First, it generates high-quality rationales by approximating the optimal thinking process through reinforcement learning, using a novel reward shaping mechanism. Second, it enhances the base LLM by maximizing the joint probability of rationale generation with respect to the model's parameters. Theoretically, we demonstrate BRiTE's convergence at a rate of $1/T$ with $T$ representing the number of iterations. Empirical evaluations on math and coding benchmarks demonstrate that our approach consistently improves performance across different base models without requiring human-annotated thinking processes. In addition, BRiTE demonstrates superior performance compared to existing algorithms that bootstrap thinking processes use alternative methods such as rejection sampling, and can even match or exceed the results achieved through supervised fine-tuning with human-annotated data.
Authors: Faming Xu, Yiding Wang, Chen Qiao, Gang Qu, Vince D. Calhoun, Julia M. Stephen, Tony W. Wilson, Yu-Ping Wang
Abstract: Dynamic effective connectivity networks (dECNs) reveal the changing directed brain activity and the dynamic causal influences among brain regions, which facilitate the identification of individual differences and enhance the understanding of human brain. Although the existing causal discovery methods have shown promising results in effective connectivity network analysis, they often overlook the dynamics of causality, in addition to the incorporation of spatio-temporal information in brain activity data. To address these issues, we propose a deep spatio-temporal fusion architecture, which employs a dynamic causal deep encoder to incorporate spatio-temporal information into dynamic causality modeling, and a dynamic causal deep decoder to verify the discovered causality. The effectiveness of the proposed method is first illustrated with simulated data. Then, experimental results from Philadelphia Neurodevelopmental Cohort (PNC) demonstrate the superiority of the proposed method in inferring dECNs, which reveal the dynamic evolution of directed flow between brain regions. The analysis shows the difference of dECNs between young adults and children. Specifically, the directed brain functional networks transit from fluctuating undifferentiated systems to more stable specialized networks as one grows. This observation provides further evidence on the modularization and adaptation of brain networks during development, leading to higher cognitive abilities observed in young adults.
Authors: Tom Overman, Diego Klabjan
Abstract: Federated averaging (FedAvg) is a popular algorithm for horizontal federated learning (FL), where samples are gathered across different clients and are not shared with each other or a central server. Extensive convergence analysis of FedAvg exists for the discrete iteration setting, guaranteeing convergence for a range of loss functions and varying levels of data heterogeneity. We extend this analysis to the continuous-time setting where the global weights evolve according to a multivariate stochastic differential equation (SDE), which is the first time FedAvg has been studied from the continuous-time perspective. We use techniques from stochastic processes to establish convergence guarantees under different loss functions, some of which are more general than existing work in the discrete setting. We also provide conditions for which FedAvg updates to the server weights can be approximated as normal random variables. Finally, we use the continuous-time formulation to reveal generalization properties of FedAvg.
Authors: Macheng Shen, Chen Cheng
Abstract: Inspired by the ubiquitous use of differential equations to model continuous dynamics across diverse scientific and engineering domains, we propose a novel and intuitive approach to continuous sequence modeling. Our method interprets time-series data as \textit{discrete samples from an underlying continuous dynamical system}, and models its time evolution using Neural Stochastic Differential Equation (Neural SDE), where both the flow (drift) and diffusion terms are parameterized by neural networks. We derive a principled maximum likelihood objective and a \textit{simulation-free} scheme for efficient training of our Neural SDE model. We demonstrate the versatility of our approach through experiments on sequence modeling tasks across both embodied and generative AI. Notably, to the best of our knowledge, this is the first work to show that SDE-based continuous-time modeling also excels in such complex scenarios, and we hope that our work opens up new avenues for research of SDE models in high-dimensional and temporally intricate domains.
Authors: Akhil Agnihotri, Rahul Jain, Deepak Ramachandran, Zheng Wen
Abstract: We address the problem of best policy identification in preference-based reinforcement learning (PbRL), where learning occurs from noisy binary preferences over trajectory pairs rather than explicit numerical rewards. This approach is useful for post-training optimization of generative AI models during multi-turn user interactions, where preference feedback is more robust than handcrafted reward models. In this setting, learning is driven by both an offline preference dataset -- collected from a rater of unknown 'competence' -- and online data collected with pure exploration. Since offline datasets may exhibit out-of-distribution (OOD) biases, principled online data collection is necessary. To address this, we propose Posterior Sampling for Preference Learning ($\mathsf{PSPL}$), a novel algorithm inspired by Top-Two Thompson Sampling, that maintains independent posteriors over the true reward model and transition dynamics. We provide the first theoretical guarantees for PbRL in this setting, establishing an upper bound on the simple Bayesian regret of $\mathsf{PSPL}$. Since the exact algorithm can be computationally impractical, we also provide an approximate version that outperforms existing baselines.
Authors: M. Hadi Sepanj, Benyamin Ghojogh, Paul Fieguth
Abstract: Self-supervised learning has gained significant attention in contemporary applications, particularly due to the scarcity of labeled data. While existing SSL methodologies primarily address feature variance and linear correlations, they often neglect the intricate relations between samples and the nonlinear dependencies inherent in complex data. In this paper, we introduce Correlation-Dependence Self-Supervised Learning (CDSSL), a novel framework that unifies and extends existing SSL paradigms by integrating both linear correlations and nonlinear dependencies, encapsulating sample-wise and feature-wise interactions. Our approach incorporates the Hilbert-Schmidt Independence Criterion (HSIC) to robustly capture nonlinear dependencies within a Reproducing Kernel Hilbert Space, enriching representation learning. Experimental evaluations on diverse benchmarks demonstrate the efficacy of CDSSL in improving representation quality.
Authors: Takeshi Koshizuka, Issei Sato
Abstract: In recent years, physics-informed machine learning has gained significant attention for its ability to enhance statistical performance and sample efficiency by integrating physical structures into machine learning models. These structures, such as differential equations, conservation laws, and symmetries, serve as inductive biases that can improve the generalization capacity of the hybrid model. However, the mechanisms by which these physical structures enhance generalization capacity are not fully understood, limiting the ability to guarantee the performance of the models. In this study, we show that the generalization performance of linear regressors incorporating differential equation structures is determined by the dimension of the associated affine variety, rather than the number of parameters. This finding enables a unified analysis of various equations, including nonlinear ones. We introduce a method to approximate the dimension of the affine variety and provide experimental evidence to validate our theoretical insights.
Authors: Shichang Zhang, Tessa Han, Usha Bhalla, Hima Lakkaraju
Abstract: The increasing complexity of AI systems has made understanding their behavior a critical challenge. Numerous methods have been developed to attribute model behavior to three key aspects: input features, training data, and internal model components. However, these attribution methods are studied and applied rather independently, resulting in a fragmented landscape of approaches and terminology. This position paper argues that feature, data, and component attribution methods share fundamental similarities, and bridging them can benefit interpretability research. We conduct a detailed analysis of successful methods across three domains and present a unified view to demonstrate that these seemingly distinct methods employ similar approaches, such as perturbations, gradients, and linear approximations, differing primarily in their perspectives rather than core techniques. Our unified perspective enhances understanding of existing attribution methods, identifies shared concepts and challenges, makes this field more accessible to newcomers, and highlights new directions not only for attribution and interpretability but also for broader AI research, including model editing, steering, and regulation.
Authors: Mohammad Al Olaimat, Serdar Bozdag
Abstract: Electronic health records (EHRs) provide a comprehensive source of longitudinal patient data, encompassing structured modalities such as laboratory results, imaging data, and vital signs, and unstructured clinical notes. These datasets, after necessary preprocessing to clean and format the data for analysis, often remain in their raw EHR form, representing numerical or categorical values without further transformation into task-agnostic embeddings. While such raw EHR data enables predictive modeling, its reliance on manual feature engineering or downstream task-specific optimization limits its utility for general-purpose applications. Deep learning (DL) techniques, such as recurrent neural networks (RNNs) and Transformers, have facilitated predictive tasks like disease progression and diagnosis prediction. However, these methods often struggle to fully exploit the temporal and multimodal dependencies inherent in EHR data due to their reliance on pre-processed but untransformed raw EHR inputs. In this study, we introduce CAAT-EHR, a novel architecture designed to bridge this gap by generating robust, task-agnostic longitudinal embeddings from raw EHR data. CAAT-EHR leverages self- and cross-attention mechanisms in its encoder to integrate temporal and contextual relationships across multiple modalities, transforming the data into enriched embeddings that capture complex dependencies. An autoregressive decoder complements the encoder by predicting future time points data during pre-training, ensuring that the resulting embeddings maintain temporal consistency and alignment. CAAT-EHR eliminates the need for manual feature engineering and enables seamless transferability across diverse downstream tasks. Extensive evaluations on benchmark datasets, demonstrate the superiority of CAAT-EHR-generated embeddings over pre-processed raw EHR data and other baseline approaches.
Authors: Mohamad Roshanzamir, Roohallah Alizadehsani, Ehsan Zarepur, Noushin Mohammadifard, Fatemeh Nouri, Mahdi Roshanzamir, Alireza Khosravi, Fereidoon Nouhi, Nizal Sarrafzadegan
Abstract: Premature coronary artery disease (PCAD) refers to the early onset of the disease, usually before the age of 55 for men and 65 for women. Coronary Artery Disease (CAD) develops when coronary arteries, the major blood vessels supplying the heart with blood, oxygen, and nutrients, become clogged or diseased. This is often due to many risk factors, including lifestyle and cardiometabolic ones, but few studies were done on ethnicity as one of these risk factors, especially in PCAD. In this study, we tested the rank of ethnicity among the major risk factors of PCAD, including age, gender, body mass index (BMI), visceral obesity presented as waist circumference (WC), diabetes mellitus (DM), high blood pressure (HBP), high low-density lipoprotein cholesterol (LDL-C), and smoking in a large national sample of patients with PCAD from different ethnicities. All patients who met the age criteria underwent coronary angiography to confirm CAD diagnosis. The weight of ethnicity was compared to the other eight features using feature weighting algorithms in PCAD diagnosis. In addition, we conducted an experiment where we ran predictive models (classification algorithms) to predict PCAD. We compared the performance of these models under two conditions: we trained the classification algorithms, including or excluding ethnicity. This study analyzed various factors to determine their predictive power influencing PCAD prediction. Among these factors, gender and age were the most significant predictors, with ethnicity being the third most important. The results also showed that if ethnicity is used as one of the input risk factors for classification algorithms, it can improve their efficiency. Our results show that ethnicity ranks as an influential factor in predicting PCAD. Therefore, it needs to be addressed in the PCAD diagnostic and preventive measures.
Authors: Khai Nguyen, Hai Nguyen, Tuan Pham, Nhat Ho
Abstract: We introduce sliced optimal transport dataset distance (s-OTDD), a model-agnostic, embedding-agnostic approach for dataset comparison that requires no training, is robust to variations in the number of classes, and can handle disjoint label sets. The core innovation is Moment Transform Projection (MTP), which maps a label, represented as a distribution over features, to a real number. Using MTP, we derive a data point projection that transforms datasets into one-dimensional distributions. The s-OTDD is defined as the expected Wasserstein distance between the projected distributions, with respect to random projection parameters. Leveraging the closed form solution of one-dimensional optimal transport, s-OTDD achieves (near-)linear computational complexity in the number of data points and feature dimensions and is independent of the number of classes. With its geometrically meaningful projection, s-OTDD strongly correlates with the optimal transport dataset distance while being more efficient than existing dataset discrepancy measures. Moreover, it correlates well with the performance gap in transfer learning and classification accuracy in data augmentation.
Authors: Ryan McKenna, Yangsibo Huang, Amer Sinha, Borja Balle, Zachary Charles, Christopher A. Choquette-Choo, Badih Ghazi, George Kaissis, Ravi Kumar, Ruibo Liu, Da Yu, Chiyuan Zhang
Abstract: Scaling laws have emerged as important components of large language model (LLM) training as they can predict performance gains through scale, and provide guidance on important hyper-parameter choices that would otherwise be expensive. LLMs also rely on large, high-quality training datasets, like those sourced from (sometimes sensitive) user data. Training models on this sensitive user data requires careful privacy protections like differential privacy (DP). However, the dynamics of DP training are significantly different, and consequently their scaling laws are not yet fully understood. In this work, we establish scaling laws that accurately model the intricacies of DP LLM training, providing a complete picture of the compute-privacy-utility tradeoffs and the optimal training configurations in many settings.
Authors: Giovanni Luca Marchetti, Vahid Shahverdi, Stefano Mereta, Matthew Trager, Kathl\'en Kohn
Abstract: In this expository work, we promote the study of function spaces parameterized by machine learning models through the lens of algebraic geometry. To this end, we focus on algebraic models, such as neural networks with polynomial activations, whose associated function spaces are semi-algebraic varieties. We outline a dictionary between algebro-geometric invariants of these varieties, such as dimension, degree, and singularities, and fundamental aspects of machine learning, such as sample complexity, expressivity, training dynamics, and implicit bias. Along the way, we review the literature and discuss ideas beyond the algebraic domain. This work lays the foundations of a research direction bridging algebraic geometry and deep learning, that we refer to as neuroalgebraic geometry.
Authors: Sagnik Anupam, Alexander Shypula, Osbert Bastani
Abstract: With the advent of large language models (LLMs), there has been a great deal of interest in applying them to solve difficult programming tasks. Recent work has demonstrated their potential at program optimization, a key challenge in programming languages research. We propose a blackbox adaptation method called Retrieval Augmented Search (RAS) that performs beam search over candidate optimizations; at each step, it retrieves in-context examples from a given training dataset of slow-fast program pairs to guide the LLM. Critically, we find that performing contextual retrieval based on an LLM-generated natural language description significantly outperforms retrieval based on the source code. In addition, we propose a method called AEGIS for improving interpretability by decomposing training examples into "atomic edits" that are significantly more incremental in nature. We show that RAS performs 1.8$\times$ better than prior state-of-the-art blackbox adaptation strategies, and that AEGIS performs 1.37$\times$ better while performing significantly smaller edits.
Authors: Zi-Jian Cheng, Zi-Yi Jia, Zhi Zhou, Lan-Zhe Guo, Yu-Feng Li
Abstract: Tabular data is widely utilized in various machine learning tasks. Current tabular learning research predominantly focuses on closed environments, while in real-world applications, open environments are often encountered, where distribution and feature shifts occur, leading to significant degradation in model performance. Previous research has primarily concentrated on mitigating distribution shifts, whereas feature shifts, a distinctive and unexplored challenge of tabular data, have garnered limited attention. To this end, this paper conducts the first comprehensive study on feature shifts in tabular data and introduces the first tabular feature-shift benchmark (TabFSBench). TabFSBench evaluates impacts of four distinct feature-shift scenarios on four tabular model categories across various datasets and assesses the performance of large language models (LLMs) and tabular LLMs in the tabular benchmark for the first time. Our study demonstrates three main observations: (1) most tabular models have the limited applicability in feature-shift scenarios; (2) the shifted feature set importance has a linear relationship with model performance degradation; (3) model performance in closed environments correlates with feature-shift performance. Future research direction is also explored for each observation. TabFSBench is released for public access by using a few lines of Python codes at https://github.com/LAMDASZ-ML/TabFSBench.
Authors: Minh Le, Anh Nguyen, Huy Nguyen, Chau Nguyen, Nhat Ho
Abstract: Visual Prompt Tuning (VPT) has recently emerged as a powerful method for adapting pre-trained vision models to downstream tasks. By introducing learnable prompt tokens as task-specific instructions, VPT effectively guides pre-trained transformer models with minimal overhead. Despite its empirical success, a comprehensive theoretical understanding of VPT remains an active area of research. Building on recent insights into the connection between mixture of experts and prompt-based approaches, we identify a key limitation in VPT: the restricted functional expressiveness in prompt formulation. To address this limitation, we propose Visual Adaptive Prompt Tuning (VAPT), a new generation of prompts that redefines prompts as adaptive functions of the input. Our theoretical analysis shows that this simple yet intuitive approach achieves optimal sample efficiency. Empirical results on VTAB-1K and FGVC further demonstrate VAPT's effectiveness, with performance gains of 7.34% and 1.04% over fully fine-tuning baselines, respectively. Notably, VAPT also surpasses VPT by a substantial margin while using fewer parameters. These results highlight both the effectiveness and efficiency of our method and pave the way for future research to explore the potential of adaptive prompts.
Authors: The Viet Bui, Tien Mai, Hong Thanh Nguyen
Abstract: Inferring reward functions from demonstrations is a key challenge in reinforcement learning (RL), particularly in multi-agent RL (MARL), where large joint state-action spaces and complex inter-agent interactions complicate the task. While prior single-agent studies have explored recovering reward functions and policies from human preferences, similar work in MARL is limited. Existing methods often involve separate stages of supervised reward learning and MARL algorithms, leading to unstable training. In this work, we introduce a novel end-to-end preference-based learning framework for cooperative MARL, leveraging the underlying connection between reward functions and soft Q-functions. Our approach uses a carefully-designed multi-agent value decomposition strategy to improve training efficiency. Extensive experiments on SMAC and MAMuJoCo benchmarks show that our algorithm outperforms existing methods across various tasks.
Authors: Anh Bui, Trang Vu, Long Vuong, Trung Le, Paul Montague, Tamas Abraham, Junae Kim, Dinh Phung
Abstract: Concept erasure has emerged as a promising technique for mitigating the risk of harmful content generation in diffusion models by selectively unlearning undesirable concepts. The common principle of previous works to remove a specific concept is to map it to a fixed generic concept, such as a neutral concept or just an empty text prompt. In this paper, we demonstrate that this fixed-target strategy is suboptimal, as it fails to account for the impact of erasing one concept on the others. To address this limitation, we model the concept space as a graph and empirically analyze the effects of erasing one concept on the remaining concepts. Our analysis uncovers intriguing geometric properties of the concept space, where the influence of erasing a concept is confined to a local region. Building on this insight, we propose the Adaptive Guided Erasure (AGE) method, which \emph{dynamically} selects optimal target concepts tailored to each undesirable concept, minimizing unintended side effects. Experimental results show that AGE significantly outperforms state-of-the-art erasure methods on preserving unrelated concepts while maintaining effective erasure performance. Our code is published at {https://github.com/tuananhbui89/Adaptive-Guided-Erasure}.
URLs: https://github.com/tuananhbui89/Adaptive-Guided-Erasure
Authors: Xin Li, Xiaotao Zheng, Zhihong Xia
Abstract: XNet is a single-layer neural network architecture that leverages Cauchy integral-based activation functions for high-order function approximation. Through theoretical analysis, we show that the Cauchy activation functions used in XNet can achieve arbitrary-order polynomial convergence, fundamentally outperforming traditional MLPs and Kolmogorov-Arnold Networks (KANs) that rely on increased depth or B-spline activations. Our extensive experiments on function approximation, PDE solving, and reinforcement learning demonstrate XNet's superior performance - reducing approximation error by up to 50000 times and accelerating training by up to 10 times compared to existing approaches. These results establish XNet as a highly efficient architecture for both scientific computing and AI applications.
Authors: Pu Yang, Yunzhen Feng, Ziyuan Chen, Yuhang Wu, Zhuoyuan Li
Abstract: Modern foundation models often undergo iterative ``bootstrapping'' in their post-training phase: a model generates synthetic data, an external verifier filters out low-quality samples, and the high-quality subset is used for further fine-tuning. Over multiple iterations, the model's performance improves--raising a crucial question: how should the total budget on generation and training be allocated across iterations to maximize final performance? In this work, we develop a theoretical framework to analyze budget allocation strategies. Specifically, we show that constant policies fail to converge with high probability, while increasing policies--particularly exponential growth policies--exhibit significant theoretical advantages. Experiments on image denoising with diffusion probabilistic models and math reasoning with large language models show that both exponential and polynomial growth policies consistently outperform constant policies, with exponential policies often providing more stable performance.
Authors: Fabian Schaipp, Alexander H\"agele, Adrien Taylor, Umut Simsekli, Francis Bach
Abstract: We show that learning-rate schedules for large model training behave surprisingly similar to a performance bound from non-smooth convex optimization theory. We provide a bound for the constant schedule with linear cooldown; in particular, the practical benefit of cooldown is reflected in the bound due to the absence of logarithmic terms. Further, we show that this surprisingly close match between optimization theory and practice can be exploited for learning-rate tuning: we achieve noticeable improvements for training 124M and 210M Llama-type models by (i) extending the schedule for continued training with optimal learning-rate, and (ii) transferring the optimal learning-rate across schedules.
Authors: Yuxuan Liu, Jingmin Sun, Hayden Schaeffer
Abstract: We introduce BCAT, a PDE foundation model designed for autoregressive prediction of solutions to two dimensional fluid dynamics problems. Our approach uses a block causal transformer architecture to model next frame predictions, leveraging previous frames as contextual priors rather than relying solely on sub-frames or pixel-based inputs commonly used in image generation methods. This block causal framework more effectively captures the spatial dependencies inherent in nonlinear spatiotemporal dynamics and physical phenomena. In an ablation study, next frame prediction demonstrated a 2.9x accuracy improvement over next token prediction. BCAT is trained on a diverse range of fluid dynamics datasets, including incompressible and compressible Navier-Stokes equations across various geometries and parameter regimes, as well as the shallow-water equations. The model's performance was evaluated on 6 distinct downstream prediction tasks and tested on about 8K trajectories to measure robustness on a variety of fluid dynamics simulations. BCAT achieved an average relative error of 1.92% across all evaluation tasks, outperforming prior approaches on standard benchmarks.
Authors: Seungheun Baek, Soyon Park, Yan Ting Chok, Mogan Gim, Jaewoo Kang
Abstract: Motivation: Predicting cellular responses to genetic perturbations is essential for understanding biological systems and developing targeted therapeutic strategies. While variational autoencoders (VAEs) have shown promise in modeling perturbation responses, their limited explainability poses a significant challenge, as the learned features often lack clear biological meaning. Nevertheless, model explainability is one of the most important aspects in the realm of biological AI. One of the most effective ways to achieve explainability is incorporating the concept of gene regulatory networks (GRNs) in designing deep learning models such as VAEs. GRNs elicit the underlying causal relationships between genes and are capable of explaining the transcriptional responses caused by genetic perturbation treatments. Results: We propose GPO-VAE, an explainable VAE enhanced by GRN-aligned Parameter Optimization that explicitly models gene regulatory networks in the latent space. Our key approach is to optimize the learnable parameters related to latent perturbation effects towards GRN-aligned explainability. Experimental results on perturbation prediction show our model achieves state-of-the-art performance in predicting transcriptional responses across multiple benchmark datasets. Furthermore, additional results on evaluating the GRN inference task reveal our model's ability to generate meaningful GRNs compared to other methods. According to qualitative analysis, GPO-VAE posseses the ability to construct biologically explainable GRNs that align with experimentally validated regulatory pathways. GPO-VAE is available at https://github.com/dmis-lab/GPO-VAE
Authors: Mathieu Even, Laurent Massouli\'e
Abstract: Motivated by multi-task and meta-learning approaches, we consider the problem of learning structure shared by tasks or users, such as shared low-rank representations or clustered structures. While all previous works focus on well-specified linear regression, we consider more general convex objectives, where the structural low-rank and cluster assumptions are expressed on the optima of each function. We show that under mild assumptions such as \textit{Hessian concentration} and \textit{noise concentration at the optimum}, rank and clustered regularized estimators recover such structure, provided the number of samples per task and the number of tasks are large enough. We then study the problem of recovering the subspace in which all the solutions lie, in the setting where there is only a single sample per task: we show that in that case, the rank-constrained estimator can recover the subspace, but that the number of tasks needs to scale exponentially large with the dimension of the subspace. Finally, we provide a polynomial-time algorithm via nuclear norm constraints for learning a shared linear representation in the context of convex learning objectives.
Authors: Kai Yi, Peter Richt\'arik
Abstract: Popular post-training pruning methods such as Wanda and RIA are known for their simple, yet effective, designs that have shown exceptional empirical performance. Wanda optimizes performance through calibrated activations during pruning, while RIA emphasizes the relative, rather than absolute, importance of weight elements. Despite their practical success, a thorough theoretical foundation explaining these outcomes has been lacking. This paper introduces new theoretical insights that redefine the standard minimization objective for pruning, offering a deeper understanding of the factors contributing to their success. Our study extends beyond these insights by proposing complementary strategies that consider both input activations and weight significance. We validate these approaches through rigorous experiments, demonstrating substantial enhancements over existing methods. Furthermore, we introduce a novel training-free fine-tuning approach $R^2$-DSnoT that incorporates relative weight importance and a regularized decision boundary within a dynamic pruning-and-growing framework, significantly outperforming strong baselines and establishing a new state of the art.
Authors: Xinshuai Dong, Ignavier Ng, Boyang Sun, Haoyue Dai, Guang-Yuan Hao, Shunxing Fan, Peter Spirtes, Yumou Qiu, Kun Zhang
Abstract: Recent advances have shown that statistical tests for the rank of cross-covariance matrices play an important role in causal discovery. These rank tests include partial correlation tests as special cases and provide further graphical information about latent variables. Existing rank tests typically assume that all the continuous variables can be perfectly measured, and yet, in practice many variables can only be measured after discretization. For example, in psychometric studies, the continuous level of certain personality dimensions of a person can only be measured after being discretized into order-preserving options such as disagree, neutral, and agree. Motivated by this, we propose Mixed data Permutation-based Rank Test (MPRT), which properly controls the statistical errors even when some or all variables are discretized. Theoretically, we establish the exchangeability and estimate the asymptotic null distribution by permutations; as a consequence, MPRT can effectively control the Type I error in the presence of discretization while previous methods cannot. Empirically, our method is validated by extensive experiments on synthetic data and real-world data to demonstrate its effectiveness as well as applicability in causal discovery.
Authors: Arjun Krishna, Erick Galinkin, Leon Derczynski, Jeffrey Martin
Abstract: Large Language Models (LLMs) have become an essential tool in the programmer's toolkit, but their tendency to hallucinate code can be used by malicious actors to introduce vulnerabilities to broad swathes of the software supply chain. In this work, we analyze package hallucination behaviour in LLMs across popular programming languages examining both existing package references and fictional dependencies. By analyzing this package hallucination behaviour we find potential attacks and suggest defensive strategies to defend against these attacks. We discover that package hallucination rate is predicated not only on model choice, but also programming language, model size, and specificity of the coding task request. The Pareto optimality boundary between code generation performance and package hallucination is sparsely populated, suggesting that coding models are not being optimized for secure code. Additionally, we find an inverse correlation between package hallucination rate and the HumanEval coding benchmark, offering a heuristic for evaluating the propensity of a model to hallucinate packages. Our metrics, findings and analyses provide a base for future models, securing AI-assisted software development workflows against package supply chain attacks.
Authors: Ahmed K. Kadhim, Lei Jiao, Rishad Shafik, Ole-Christoffer Granmo, Bimal Bhattarai
Abstract: The Tsetlin Machine (TM) architecture has recently demonstrated effectiveness in Machine Learning (ML), particularly within Natural Language Processing (NLP). It has been utilized to construct word embedding using conjunctive propositional clauses, thereby significantly enhancing our understanding and interpretation of machine-derived decisions. The previous approach performed the word embedding over a sequence of input words to consolidate the information into a cohesive and unified representation. However, that approach encounters scalability challenges as the input size increases. In this study, we introduce a novel approach incorporating two-phase training to discover contextual embeddings of input sequences. Specifically, this method encapsulates the knowledge for each input word within the dataset's vocabulary, subsequently constructing embeddings for a sequence of input words utilizing the extracted knowledge. This technique not only facilitates the design of a scalable model but also preserves interpretability. Our experimental findings revealed that the proposed method yields competitive performance compared to the previous approaches, demonstrating promising results in contrast to human-generated benchmarks. Furthermore, we applied the proposed approach to sentiment analysis on the IMDB dataset, where the TM embedding and the TM classifier, along with other interpretable classifiers, offered a transparent end-to-end solution with competitive performance.
Authors: Abdulrahman Altahhan
Abstract: In this paper, we develop a new planning method that extends the capabilities of the true online TD to allow an agent to efficiently replay all or part of its past experience, online in the sequence that they appear with, either in each step or sparsely according to the usual {\lambda} parameter. In this new method that we call True Online TD-Replan({\lambda}), the {\lambda} parameter plays a new role in specifying the density of the replay process in addition to the usual role of specifying the depth of the target's updates. We demonstrate that, for problems that benefit from experience replay, our new method outperforms true online TD({\lambda}), albeit quadratic in complexity due to its replay capabilities. In addition, we demonstrate that our method outperforms other methods with similar quadratic complexity such as Dyna Planning and TD({\lambda})-Replan algorithms. We test our method on two benchmarking environments, a random walk problem that uses simple binary features and a myoelectric control domain that uses both simple sEMG features and deeply extracted features to showcase its capabilities.
Authors: Han Yu, Jiashuo Liu, Hao Zou, Renzhe Xu, Yue He, Xingxuan Zhang, Peng Cui
Abstract: Despite the great performance of deep learning models in many areas, they still make mistakes and underperform on certain subsets of data, i.e. error slices. Given a trained model, it is important to identify its semantically coherent error slices that are easy to interpret, which is referred to as the error slice discovery problem. However, there is no proper metric of slice coherence without relying on extra information like predefined slice labels. Current evaluation of slice coherence requires access to predefined slices formulated by metadata like attributes or subclasses. Its validity heavily relies on the quality and abundance of metadata, where some possible patterns could be ignored. Besides, current algorithms cannot directly incorporate the constraint of coherence into their optimization objective due to the absence of an explicit coherence metric, which could potentially hinder their effectiveness. In this paper, we propose manifold compactness, a coherence metric without reliance on extra information by incorporating the data geometry property into its design, and experiments on typical datasets empirically validate the rationality of the metric. Then we develop Manifold Compactness based error Slice Discovery (MCSD), a novel algorithm that directly treats risk and coherence as the optimization objective, and is flexible to be applied to models of various tasks. Extensive experiments on the benchmark and case studies on other typical datasets demonstrate the superiority of MCSD.
Authors: Huanran Chen, Yinpeng Dong, Zeming Wei, Hang Su, Jun Zhu
Abstract: Recent studies have revealed the vulnerability of Large Language Models (LLMs) to adversarial attacks, where the adversary crafts specific input sequences to induce harmful, violent, private, or incorrect outputs. Although various defenses have been proposed, they have not been evaluated by strong adaptive attacks, leaving the worst-case robustness of LLMs still intractable. By developing a stronger white-box attack, our evaluation results indicate that most typical defenses achieve nearly 0\% robustness.To solve this, we propose \textit{DiffTextPure}, a general defense that diffuses the (adversarial) input prompt using any pre-defined smoothing distribution, and purifies the diffused input using a pre-trained language model. Theoretically, we derive tight robustness lower bounds for all smoothing distributions using Fractal Knapsack or 0-1 Knapsack solvers. Under this framework, we certify the robustness of a specific case -- smoothing LLMs using a uniform kernel -- against \textit{any possible attack} with an average $\ell_0$ perturbation of 2.02 or an average suffix length of 6.41.
Authors: Ruigang Wang, Krishnamurthy Dvijotham, Ian R. Manchester
Abstract: In this work, we propose norm-bounded low-rank adaptation (NB-LoRA) for parameter-efficient fine tuning. We introduce two parameterizations that allow explicit bounds on each singular value of the weight adaptation matrix, which can therefore satisfy any prescribed unitarily invariant norm bound, including the Schatten norms (e.g., nuclear, Frobenius, spectral norm). The proposed parameterizations are unconstrained and complete, i.e. they cover all matrices satisfying the prescribed rank and norm constraints. Experiments on vision fine-tuning benchmarks show that the proposed approach can achieve good adaptation performance while avoiding model catastrophic forgetting and also substantially improve robustness to a wide range of hyper-parameters, including adaptation rank, learning rate and number of training epochs. We also explore applications in privacy-preserving model merging and low-rank matrix completion.
Authors: Lingwei Zhu, Zheng Chen, Yukie Nagai, Jimeng Sun
Abstract: This paper adds to the growing literature of reinforcement learning (RL) for healthcare by proposing a novel paradigm: augmenting any predictor with Rule-based RL Layer (RRLL) that corrects the model's physiologically impossible predictions. Specifically, RRLL takes as input states predicted labels and outputs corrected labels as actions. The reward of the state-action pair is evaluated by a set of general rules. RRLL is efficient, general and lightweight: it does not require heavy expert knowledge like prior work but only a set of impossible transitions. This set is much smaller than all possible transitions; yet it can effectively reduce physiologically impossible mistakes made by the state-of-the-art predictor models. We verify the utility of RRLL on a variety of important healthcare classification problems and observe significant improvements using the same setup, with only the domain-specific set of impossibility changed. In-depth analysis shows that RRLL indeed improves accuracy by effectively reducing the presence of physiologically impossible predictions.
Authors: Yan Sun, Tiansheng Huang, Liang Ding, Li Shen, Dacheng Tao
Abstract: Zeroth-order optimization (ZO) has demonstrated remarkable promise in efficient fine-tuning tasks for Large Language Models (LLMs). In particular, recent advances incorporate the low-rankness of gradients, introducing low-rank ZO estimators to further reduce GPU memory consumption. However, most existing works focus solely on the low-rankness of each individual gradient, overlooking a broader property shared by all gradients throughout the training, i.e., all gradients approximately reside within a similar subspace. In this paper, we consider two factors together and propose a novel low-rank ZO estimator, TeZO, which captures the low-rankness across both the model and temporal dimension. Specifically, we represent ZO perturbations along the temporal dimension as a 3D tensor and employ Canonical Polyadic Decomposition (CPD) to extract each low-rank 2D matrix, significantly reducing the training cost. TeZO can also be easily extended to the Adam variant while consuming less memory than MeZO-SGD, and requiring about only 35% memory of MeZO-Adam. Both comprehensive theoretical analysis and extensive experimental research have validated its efficiency, achieving SOTA-comparable results with lower overhead of time and memory.
Authors: Lars C. P. M. Quaedvlieg
Abstract: Efficient job allocation in complex scheduling problems poses significant challenges in real-world applications. In this report, we propose a novel approach that leverages the power of Reinforcement Learning (RL) and Graph Neural Networks (GNNs) to tackle the Job Allocation Problem (JAP). The JAP involves allocating a maximum set of jobs to available resources while considering several constraints. Our approach enables learning of adaptive policies through trial-and-error interactions with the environment while exploiting the graph-structured data of the problem. By leveraging RL, we eliminate the need for manual annotation, a major bottleneck in supervised learning approaches. Experimental evaluations on synthetic and real-world data demonstrate the effectiveness and generalizability of our proposed approach, outperforming baseline algorithms and showcasing its potential for optimizing job allocation in complex scheduling problems.
Authors: Zhixuan Li, Naipeng Chen, Seonghwa Choi, Sanghoon Lee, Weisi Lin
Abstract: Time-series forecasting is crucial for numerous real-world applications including weather prediction and financial market modeling. While temporal-domain methods remain prevalent, frequency-domain approaches can effectively capture multi-scale periodic patterns, reduce sequence dependencies, and naturally denoise signals. However, existing approaches typically train model components for all frequencies under a unified training objective, often leading to mismatched learning speeds: high-frequency components converge faster and risk overfitting, while low-frequency components underfit due to insufficient training time. To deal with this challenge, we propose BEAT (Balanced frEquency Adaptive Tuning), a novel framework that dynamically monitors the training status for each frequency and adaptively adjusts their gradient updates. By recognizing convergence, overfitting, or underfitting for each frequency, BEAT dynamically reallocates learning priorities, moderating gradients for rapid learners and increasing those for slower ones, alleviating the tension between competing objectives across frequencies and synchronizing the overall learning process. Extensive experiments on seven real-world datasets demonstrate that BEAT consistently outperforms state-of-the-art approaches.
Authors: Hossein Zakerinia, Dorsa Ghobadi, Christoph H. Lampert
Abstract: Deep learning methods are known to generalize well from training to future data, even in an overparametrized regime, where they could easily overfit. One explanation for this phenomenon is that even when their *ambient dimensionality*, (i.e. the number of parameters) is large, the models' *intrinsic dimensionality* is small, i.e. their learning takes place in a small subspace of all possible weight configurations. In this work, we confirm this phenomenon in the setting of *deep multi-task learning*. We introduce a method to parametrize multi-task network directly in the low-dimensional space, facilitated by the use of *random expansions* techniques. We then show that high-accuracy multi-task solutions can be found with much smaller intrinsic dimensionality (fewer free parameters) than what single-task learning requires. Subsequently, we show that the low-dimensional representations in combination with *weight compression* and *PAC-Bayesian* reasoning lead to the first *non-vacuous generalization bounds* for deep multi-task networks.
Authors: Masanori Ishikura, Masayuki Karasuyama
Abstract: This study considers multi-objective Bayesian optimization (MOBO) through the information gain of the Pareto-frontier. To calculate the information gain, a predictive distribution conditioned on the Pareto-frontier plays a key role, which is defined as a distribution truncated by the Pareto-frontier. However, it is usually impossible to obtain the entire Pareto-frontier in a continuous domain, and therefore, the complete truncation cannot be known. We consider an approximation of the truncate distribution by using a mixture distribution consisting of two possible approximate truncation obtainable from a subset of the Pareto-frontier, which we call over- and under-truncation. Since the optimal balance of the mixture is unknown beforehand, we propose optimizing the balancing coefficient through the variational lower bound maximization framework, by which the approximation error of the information gain can be minimized. Our empirical evaluation demonstrates the effectiveness of the proposed method particularly when the number of objective functions is large.
Authors: Henrik Schopmans, Pascal Friederich
Abstract: Efficient sampling of unnormalized probability densities such as the Boltzmann distribution of molecular systems is a longstanding challenge. Next to conventional approaches like molecular dynamics or Markov chain Monte Carlo, variational approaches, such as training normalizing flows with the reverse Kullback-Leibler divergence, have been introduced. However, such methods are prone to mode collapse and often do not learn to sample the full configurational space. Here, we present temperature-annealed Boltzmann generators (TA-BG) to address this challenge. First, we demonstrate that training a normalizing flow with the reverse Kullback-Leibler divergence at high temperatures is possible without mode collapse. Furthermore, we introduce a reweighting-based training objective to anneal the distribution to lower target temperatures. We apply this methodology to three molecular systems of increasing complexity and, compared to the baseline, achieve better results in almost all metrics while requiring up to three times fewer target energy evaluations. For the largest system, our approach is the only method that accurately resolves the metastable states of the system.
Authors: Alexandre Rio, Merwan Barlier, Igor Colin
Abstract: Motivated by the increasing deployment of reinforcement learning in the real world, involving a large consumption of personal data, we introduce a differentially private (DP) policy gradient algorithm. We show that, in this setting, the introduction of Differential Privacy can be reduced to the computation of appropriate trust regions, thus avoiding the sacrifice of theoretical properties of the DP-less methods. Therefore, we show that it is possible to find the right trade-off between privacy noise and trust-region size to obtain a performant differentially private policy gradient algorithm. We then outline its performance empirically on various benchmarks. Our results and the complexity of the tasks addressed represent a significant improvement over existing DP algorithms in online RL.
Authors: Yuchen Hu, Xi Chen, Weidong Liu, Xiaojun Mao
Abstract: Distributed stochastic optimization algorithms can handle large-scale data simultaneously and accelerate model training. However, the sparsity of distributed networks and the heterogeneity of data limit these advantages. This paper proposes a momentum-accelerated distributed stochastic gradient algorithm, referred to as Exact-Diffusion with Momentum (EDM), which can correct the bias caused by data heterogeneity and introduces the momentum method commonly used in deep learning to accelerate the convergence of the algorithm. We theoretically demonstrate that this algorithm converges to the neighborhood of the optimum sub-linearly irrelevant to data heterogeneity when applied to non-convex objective functions and linearly under the Polyak-{\L}ojasiewicz condition (a weaker assumption than $\mu$-strongly convexity). Finally, we evaluate the performance of the proposed algorithm by simulation, comparing it with a range of existing decentralized optimization algorithms to demonstrate its effectiveness in addressing data heterogeneity and network sparsity.
Authors: Keqin Wang, Yulong Yang, Ishan Saha, Christine Allen-Blanchette
Abstract: In contrast to classes of neural networks where the learned representations become increasingly expressive with network depth, the learned representations in graph neural networks (GNNs), tend to become increasingly similar. This phenomena, known as oversmoothing, is characterized by learned representations that cannot be reliably differentiated leading to reduced predictive performance. In this paper, we propose an analogy between oversmoothing in GNNs and consensus or agreement in opinion dynamics. Through this analogy, we show that the message passing structure of recent continuous-depth GNNs is equivalent to a special case of opinion dynamics (i.e., linear consensus models) which has been theoretically proven to converge to consensus (i.e., oversmoothing) for all inputs. Using the understanding developed through this analogy, we design a new continuous-depth GNN model based on nonlinear opinion dynamics and prove that our model, which we call behavior-inspired message passing neural network (BIMP) circumvents oversmoothing for general inputs. Through extensive experiments, we show that BIMP is robust to oversmoothing and adversarial attack, and consistently outperforms competitive baselines on numerous benchmarks.
Authors: Jialin Zhao, Yingtao Zhang, Carlo Vittorio Cannistraci
Abstract: The rapid growth of Large Language Models has driven demand for effective model compression techniques to reduce memory and computation costs. Low-rank pruning has gained attention for its tensor coherence and GPU compatibility across all densities. However, low-rank pruning has struggled to match the performance of semi-structured pruning, often doubling perplexity (PPL) at similar densities. In this paper, we propose Pivoting Factorization (PIFA), a novel lossless meta low-rank representation that unsupervisedly learns a compact form of any low-rank representation, effectively eliminating redundant information. PIFA identifies pivot rows (linearly independent rows) and expresses non-pivot rows as linear combinations, achieving an additional 24.2\% memory savings and 24.6\% faster inference over low-rank layers at r/d = 0.5, thereby significantly enhancing performance at the same density. To mitigate the performance degradation caused by low-rank pruning, we introduce a novel, retraining-free low-rank reconstruction method that minimizes error accumulation (M). MPIFA, combining M and PIFA into an end-to-end framework, significantly outperforms existing low-rank pruning methods and, for the first time, achieves performance comparable to semi-structured pruning, while surpassing it in GPU efficiency and compatibility.
Authors: Sihwan Park, Jihun Yun, SungYub Kim, Souvik Kundu, Eunho Yang
Abstract: Zeroth-order (ZO) optimization has emerged as a promising alternative to gradient-based backpropagation methods, particularly for black-box optimization and large language model (LLM) fine-tuning. However, ZO methods suffer from slow convergence due to high-variance stochastic gradient estimators. While structured perturbations, such as sparsity and low-rank constraints, have been explored to mitigate these issues, their effectiveness remains highly under-explored. In this work, we develop a unified theoretical framework that analyzes both the convergence and generalization properties of ZO optimization under structured perturbations. We show that high dimensionality is the primary bottleneck and introduce the notions of \textit{stable rank} and \textit{effective overlap} to explain how structured perturbations reduce gradient noise and accelerate convergence. Using the uniform stability under our framework, we then provide the first theoretical justification for why these perturbations enhance generalization. Additionally, through empirical analysis, we identify that \textbf{block coordinate descent} (BCD) to be an effective structured perturbation method. Extensive experiments show that, compared to existing alternatives, memory-efficient ZO (MeZO) with BCD (\textit{MeZO-BCD}) can provide improved converge with a faster wall-clock time/iteration by up to $\times\textbf{2.09}$ while yielding similar or better accuracy.
Authors: Giulio Masinelli, Chang Rajani, Patrik Hoffmann, Kilian Wasmer, David Atienza
Abstract: Ensuring consistent processing quality is challenging in laser processes due to varying material properties and surface conditions. Although some approaches have shown promise in solving this problem via automation, they often rely on predetermined targets or are limited to simulated environments. To address these shortcomings, we propose a novel real-time reinforcement learning approach for laser process control, implemented on a Field Programmable Gate Array to achieve real-time execution. Our experimental results from laser welding tests on stainless steel samples with a range of surface roughnesses validated the method's ability to adapt autonomously, without relying on reward engineering or prior setup information. Specifically, the algorithm learned the correct power profile for each unique surface characteristic, demonstrating significant improvements over hand-engineered optimal constant power strategies -- up to 23% better performance on rougher surfaces and 7% on mixed surfaces. This approach represents a significant advancement in automating and optimizing laser processes, with potential applications across multiple industries.
Authors: Diyuan Wu, Marco Mondelli
Abstract: Neural Collapse is a phenomenon where the last-layer representations of a well-trained neural network converge to a highly structured geometry. In this paper, we focus on its first (and most basic) property, known as NC1: the within-class variability vanishes. While prior theoretical studies establish the occurrence of NC1 via the data-agnostic unconstrained features model, our work adopts a data-specific perspective, analyzing NC1 in a three-layer neural network, with the first two layers operating in the mean-field regime and followed by a linear layer. In particular, we establish a fundamental connection between NC1 and the loss landscape: we prove that points with small empirical loss and gradient norm (thus, close to being stationary) approximately satisfy NC1, and the closeness to NC1 is controlled by the residual loss and gradient norm. We then show that (i) gradient flow on the mean squared error converges to NC1 solutions with small empirical loss, and (ii) for well-separated data distributions, both NC1 and vanishing test loss are achieved simultaneously. This aligns with the empirical observation that NC1 emerges during training while models attain near-zero test error. Overall, our results demonstrate that NC1 arises from gradient training due to the properties of the loss landscape, and they show the co-occurrence of NC1 and small test error for certain data distributions.
Authors: Abhijeet Mulgund, Chirag Pabbaraju
Abstract: The paradigm of weak-to-strong generalization constitutes the training of a strong AI model on data labeled by a weak AI model, with the goal that the strong model nevertheless outperforms its weak supervisor on the target task of interest. For the setting of real-valued regression with the squared loss, recent work quantitatively characterizes the gain in performance of the strong model over the weak model in terms of the misfit between the strong and weak model. We generalize such a characterization to learning tasks whose loss functions correspond to arbitrary Bregman divergences when the strong class is convex. This extends the misfit-based characterization of performance gain in weak-to-strong generalization to classification tasks, as the cross-entropy loss can be expressed in terms of a Bregman divergence. In most practical scenarios, however, the strong model class may not be convex. We therefore weaken this assumption and study weak-to-strong generalization for convex combinations of $k$ strong models in the strong class, in the concrete setting of classification. This allows us to obtain a similar misfit-based characterization of performance gain, upto an additional error term that vanishes as $k$ gets large. Our theoretical findings are supported by thorough experiments on synthetic as well as real-world datasets.
Authors: Yingtao Zhang, Jialin Zhao, Wenjing Wu, Ziheng Liao, Umberto Michieli, Carlo Vittorio Cannistraci
Abstract: This study aims to enlarge our current knowledge on application of brain-inspired network science principles for training artificial neural networks (ANNs) with sparse connectivity. Dynamic sparse training (DST) can reduce the computational demands in ANNs, but faces difficulties to keep peak performance at high sparsity levels. The Cannistraci-Hebb training (CHT) is a brain-inspired method for growing connectivity in DST. CHT leverages a gradient-free, topology-driven link regrowth, which has shown ultra-sparse (1% connectivity or lower) advantage across various tasks compared to fully connected networks. Yet, CHT suffers two main drawbacks: (i) its time complexity is O(Nd^3) - N node network size, d node degree - hence it can apply only to ultra-sparse networks. (ii) it selects top link prediction scores, which is inappropriate for the early training epochs, when the network presents unreliable connections. We propose a GPU-friendly approximation of the CH link predictor, which reduces the computational complexity to O(N^3), enabling a fast implementation of CHT in large-scale models. We introduce the Cannistraci-Hebb training soft rule (CHTs), which adopts a strategy for sampling connections in both link removal and regrowth, balancing the exploration and exploitation of network topology. To improve performance, we integrate CHTs with a sigmoid gradual density decay (CHTss). Empirical results show that, using 1% of connections, CHTs outperforms fully connected networks in MLP on visual classification tasks, compressing some networks to < 30% nodes. Using 5% of the connections, CHTss outperforms fully connected networks in two Transformer-based machine translation tasks. Using 30% of the connections, CHTss achieves superior performance compared to other dynamic sparse training methods in language modeling, and it surpasses the fully connected counterpart in zero-shot evaluations.
Authors: Nhan Phan, Thu Nguyen, P{\aa}l Halvorsen, Michael A. Riegler
Abstract: Principal Component Analysis (PCA) is a commonly used tool for dimension reduction and denoising. Therefore, it is also widely used on the data prior to training a neural network. However, this approach can complicate the explanation of explainable AI (XAI) methods for the decision of the model. In this work, we analyze the potential issues with this approach and propose Principal Components-based Initialization (PCsInit), a strategy to incorporate PCA into the first layer of a neural network via initialization of the first layer in the network with the principal components, and its two variants PCsInit-Act and PCsInit-Sub. Explanations using these strategies are as direct and straightforward as for neural networks and are simpler than using PCA prior to training a neural network on the principal components. Moreover, as will be illustrated in the experiments, such training strategies can also allow further improvement of training via backpropagation.
Authors: Gaspard Lambrechts, Damien Ernst, Aditya Mahajan
Abstract: In reinforcement learning for partially observable environments, many successful algorithms were developed within the asymmetric learning paradigm. This paradigm leverages additional state information available at training time for faster learning. Although the proposed learning objectives are usually theoretically sound, these methods still lack a theoretical justification for their potential benefits. We propose such a justification for asymmetric actor-critic algorithms with linear function approximators by adapting a finite-time convergence analysis to this setting. The resulting finite-time bound reveals that the asymmetric critic eliminates an error term arising from aliasing in the agent state.
Authors: Hong Huang, Hai Yang, Yuan Chen, Jiaxun Ye, Dapeng Wu
Abstract: Federated Learning (FL) enables collaborative model training across distributed clients without data sharing, but its high computational and communication demands strain resource-constrained devices. While existing methods use dynamic pruning to improve efficiency by periodically adjusting sparse model topologies while maintaining sparsity, these approaches suffer from issues such as greedy adjustments, unstable topologies, and communication inefficiency, resulting in less robust models and suboptimal performance under data heterogeneity and partial client availability. To address these challenges, we propose Federated Robust pruning via combinatorial Thompson Sampling (FedRTS), a novel framework designed to develop robust sparse models. FedRTS enhances robustness and performance through its Thompson Sampling-based Adjustment (TSAdj) mechanism, which uses probabilistic decisions informed by stable, farsighted information instead of deterministic decisions reliant on unstable and myopic information in previous methods. Extensive experiments demonstrate that FedRTS achieves state-of-the-art performance in computer vision and natural language processing tasks while reducing communication costs, particularly excelling in scenarios with heterogeneous data distributions and partial client participation. Our codes are available at: https://github.com/Little0o0/FedRTS
Authors: Wenyun Li, Wenjie Huang
Abstract: In many real-world scenarios, reward signal for agents are exceedingly sparse, making it challenging to learn an effective reward function for reward shaping. To address this issue, our approach performs reward shaping not only by utilizing non-zero-reward transitions but also by employing the Semi-Supervised Learning (SSL) technique combined with a novel data augmentation to learn trajectory space representations from the majority of transitions, zero-reward transitions, thereby improving the efficacy of reward shaping. Experimental results in Atari and robotic manipulation demonstrate that our method effectively generalizes reward shaping to sparse reward scenarios, achieving up to four times better performance in reaching higher best scores compared to curiosity-driven methods. The proposed double entropy data augmentation enhances performance, showcasing a 15.8\% increase in best score over other augmentation methods.
Authors: Burcu K\"u\c{c}\"uko\u{g}lu, Sander Dalm, Marcel van Gerven
Abstract: The effectiveness of credit assignment in reinforcement learning (RL) when dealing with high-dimensional data is influenced by the success of representation learning via deep neural networks, and has implications for the sample efficiency of deep RL algorithms. Input decorrelation has been previously introduced as a method to speed up optimization in neural networks, and has proven impactful in both efficient deep learning and as a method for effective representation learning for deep RL algorithms. We propose a novel approach to online decorrelation in deep RL based on the decorrelated backpropagation algorithm that seamlessly integrates the decorrelation process into the RL training pipeline. Decorrelation matrices are added to each layer, which are updated using a separate decorrelation learning rule that minimizes the total decorrelation loss across all layers, in parallel to minimizing the usual RL loss. We used our approach in combination with the soft actor-critic (SAC) method, which we refer to as decorrelated soft actor-critic (DSAC). Experiments on the Atari 100k benchmark with DSAC shows, compared to the regular SAC baseline, faster training in five out of the seven games tested and improved reward performance in two games with around 50% reduction in wall-clock time, while maintaining performance levels on the other games. These results demonstrate the positive impact of network-wide decorrelation in deep RL for speeding up its sample efficiency through more effective credit assignment.
Authors: Alex O. Davies, Nirav S. Ajmeri, Telmo de Menezes e Silva Filho
Abstract: Graph learning on molecules makes use of information from both the molecular structure and the features attached to that structure. Much work has been conducted on biasing either towards structure or features, with the aim that bias bolsters performance. Identifying which information source a dataset favours, and therefore how to approach learning that dataset, is an open issue. Here we propose Noise-Noise Ratio Difference (NNRD), a quantitative metric for whether there is more useful information in structure or features. By employing iterative noising on features and structure independently, leaving the other intact, NNRD measures the degradation of information in each. We employ NNRD over a range of molecular tasks, and show that it corresponds well to a loss of information, with intuitive results that are more expressive than simple performance aggregates. Our future work will focus on expanding data domains, tasks and types, as well as refining our choice of baseline model.
Authors: Ning Chen, Shen-Huan Lyu, Tian-Shuang Wu, Yanyan Wang, Bin Tang
Abstract: In multi-label learning, leveraging contrastive learning to learn better representations faces a key challenge: selecting positive and negative samples and effectively utilizing label information. Previous studies selected positive and negative samples based on the overlap between labels and used them for label-wise loss balancing. However, these methods suffer from a complex selection process and fail to account for the varying importance of different labels. To address these problems, we propose a novel method that improves multi-label contrastive learning through label distribution. Specifically, when selecting positive and negative samples, we only need to consider whether there is an intersection between labels. To model the relationships between labels, we introduce two methods to recover label distributions from logical labels, based on Radial Basis Function (RBF) and contrastive loss, respectively. We evaluate our method on nine widely used multi-label datasets, including image and vector datasets. The results demonstrate that our method outperforms state-of-the-art methods in six evaluation metrics.
Authors: Enric Boix-Adsera
Abstract: We compute the minimum-norm weights of a deep linear ResNet, and find that the inductive bias of this architecture lies between minimizing nuclear norm and rank. This implies that, with appropriate hyperparameters, deep nonlinear ResNets have an inductive bias towards minimizing bottleneck rank.
Authors: Morgan Thomas, Albert Bou, Gianni De Fabritiis
Abstract: Chemical language models for molecular design have the potential to find solutions to multi-parameter optimization problems in drug discovery via reinforcement learning (RL). A key requirement to achieve this is the capacity to "search" chemical space to identify all molecules of interest. Here, we propose a challenging new benchmark to discover dissimilar molecules that possess similar bioactivity, a common scenario in drug discovery, but a hard problem to optimize. We show that a population of RL agents can solve the benchmark, while a single agent cannot. We also find that cooperative strategies are not significantly better than independent agents. Moreover, the performance on the benchmark scales log-linearly with the number of independent agents, showing a test-time training scaling law for chemical language models.
Authors: Giovanni Catania, Aur\'elien Decelle, Cyril Furtlehner, Beatriz Seoane
Abstract: We investigate the impact of limited data on training pairwise energy-based models for inverse problems aimed at identifying interaction networks. Utilizing the Gaussian model as testbed, we dissect training trajectories across the eigenbasis of the coupling matrix, exploiting the independent evolution of eigenmodes and revealing that the learning timescales are tied to the spectral decomposition of the empirical covariance matrix. We see that optimal points for early stopping arise from the interplay between these timescales and the initial conditions of training. Moreover, we show that finite data corrections can be accurately modeled through asymptotic random matrix theory calculations and provide the counterpart of generalized cross-validation in the energy based model context. Our analytical framework extends to binary-variable maximum-entropy pairwise models with minimal variations. These findings offer strategies to control overfitting in discrete-variable models through empirical shrinkage corrections, improving the management of overfitting in energy-based generative models.
Authors: Ali Momeni, Stefan Uhlich, Arun Venkitaraman, Chia-Yu Hsieh, Andrea Bonetti, Ryoga Matsuo, Eisaku Ohbuchi, Lorenzo Servadei
Abstract: In physics and engineering, many processes are modeled using non-differentiable black-box simulators, making the optimization of such functions particularly challenging. To address such cases, inspired by the Gradient Theorem, we propose locality-aware surrogate models for active model-based black-box optimization. We first establish a theoretical connection between gradient alignment and the minimization of a Gradient Path Integral Equation (GradPIE) loss, which enforces consistency of the surrogate's gradients in local regions of the design space. Leveraging this theoretical insight, we develop a scalable training algorithm that minimizes the GradPIE loss, enabling both offline and online learning while maintaining computational efficiency. We evaluate our approach on three real-world tasks - spanning automated in silico experiments such as coupled nonlinear oscillators, analog circuits, and optical systems - and demonstrate consistent improvements in optimization efficiency under limited query budgets. Our results offer dependable solutions for both offline and online optimization tasks where reliable gradient estimation is needed.
Authors: Georgia Channing, Aqib Mahfuz, Mark van der Wilk, Philip Torr, Fabio Pizzati, Christian Schroeder de Witt
Abstract: Recent advances in AI-generated steganography highlight its potential for safeguarding the privacy of vulnerable democratic actors, including aid workers, journalists, and whistleblowers operating in oppressive regimes. In this work, we address current limitations and establish the foundations for large-throughput generative steganography. We introduce a novel approach that enables secure and efficient steganography within latent diffusion models. We show empirically that our methods perform well across a variety of open-source latent diffusion models, particularly in generative image and video tasks.
Authors: Luka Kova\v{c}evi\'c, Thomas Gaudelet, James Opzoomer, Hagen Triendl, John Whittaker, Caroline Uhler, Lindsay Edwards, Jake P. Taylor-King
Abstract: Despite substantial efforts, deep learning has not yet delivered a transformative impact on elucidating regulatory biology, particularly in the realm of predicting gene expression profiles. Here, we argue that genuine "foundation models" of regulatory biology will remain out of reach unless guided by frameworks that integrate mechanistic insight with principled experimental design. We present one such ground-up, semi-mechanistic framework that unifies perturbation-based experimental designs across both in vitro and in vivo CRISPR screens, accounting for differentiating and non-differentiating cellular systems. By revealing previously unrecognised assumptions in published machine learning methods, our approach clarifies links with popular techniques such as variational autoencoders and structural causal models. In practice, this framework suggests a modified loss function that we demonstrate can improve predictive performance, and further suggests an error analysis that informs batching strategies. Ultimately, since cellular regulation emerges from innumerable interactions amongst largely uncharted molecular components, we contend that systems-level understanding cannot be achieved through structural biology alone. Instead, we argue that real progress will require a first-principles perspective on how experiments capture biological phenomena, how data are generated, and how these processes can be reflected in more faithful modelling architectures.
Authors: Rafael Elberg, Mircea Petrache, Denis Parra
Abstract: Compositionality and compositional generalization--the ability to understand novel combinations of known concepts--are central characteristics of human language and are hypothesized to be essential for human cognition. In machine learning, the emergence of this property has been studied in a communication game setting, where independent agents (a sender and a receiver) converge to a shared encoding policy from a set of states to a space of discrete messages, where the receiver can correctly reconstruct the states observed by the sender using only the sender's messages. The use of communication games in generation tasks is still largely unexplored, with recent methods for compositional generation focusing mainly on the use of supervised guidance (either through class labels or text). In this work, we take the first steps to fill this gap, and we present a self-supervised generative communication game-based framework for creating compositional encodings in learned representations from pre-trained encoder-decoder models. In an Iterated Learning (IL) protocol involving a sender and a receiver, we apply alternating pressures for compression and diversity of encoded discrete messages, so that the protocol converges to an efficient but unambiguous encoding. Approximate message entropy regularization is used to favor compositional encodings. Our framework is based on rigorous justifications and proofs of defining and balancing the concepts of Eficiency, Unambiguity and Non-Holisticity in encoding. We test our method on the compositional image dataset Shapes3D, demonstrating robust performance in both reconstruction and compositionality metrics, surpassing other tested discrete message frameworks.
Authors: Felix Dangel, Runa Eschenhagen, Weronika Ormaniec, Andres Fernandez, Lukas Tatzel, Agustinus Kristiadi
Abstract: Structured large matrices are prevalent in machine learning. A particularly important class is curvature matrices like the Hessian, which are central to understanding the loss landscape of neural nets (NNs), and enable second-order optimization, uncertainty quantification, model pruning, data attribution, and more. However, curvature computations can be challenging due to the complexity of automatic differentiation, and the variety and structural assumptions of curvature proxies, like sparsity and Kronecker factorization. In this position paper, we argue that linear operators -- an interface for performing matrix-vector products -- provide a general, scalable, and user-friendly abstraction to handle curvature matrices. To support this position, we developed $\textit{curvlinops}$, a library that provides curvature matrices through a unified linear operator interface. We demonstrate with $\textit{curvlinops}$ how this interface can hide complexity, simplify applications, be extensible and interoperable with other libraries, and scale to large NNs.
Authors: Eug\`ene Berta, David Holzm\"uller, Michael I. Jordan, Francis Bach
Abstract: Machine learning classifiers often produce probabilistic predictions that are critical for accurate and interpretable decision-making in various domains. The quality of these predictions is generally evaluated with proper losses like cross-entropy, which decompose into two components: calibration error assesses general under/overconfidence, while refinement error measures the ability to distinguish different classes. In this paper, we provide theoretical and empirical evidence that these two errors are not minimized simultaneously during training. Selecting the best training epoch based on validation loss thus leads to a compromise point that is suboptimal for both calibration error and, most importantly, refinement error. To address this, we introduce a new metric for early stopping and hyperparameter tuning that makes it possible to minimize refinement error during training. The calibration error is minimized after training, using standard techniques. Our method integrates seamlessly with any architecture and consistently improves performance across diverse classification tasks.
Authors: Lea Bogensperger, Dominik Narnhofer, Ahmed Allam, Konrad Schindler, Michael Krauthammer
Abstract: The goal of protein fitness optimization is to discover new protein variants with enhanced fitness for a given use. The vast search space and the sparsely populated fitness landscape, along with the discrete nature of protein sequences, pose significant challenges when trying to determine the gradient towards configurations with higher fitness. We introduce Variational Latent Generative Protein Optimization (VLGPO), a variational perspective on fitness optimization. Our method embeds protein sequences in a continuous latent space to enable efficient sampling from the fitness distribution and combines a (learned) flow matching prior over sequence mutations with a fitness predictor to guide optimization towards sequences with high fitness. VLGPO achieves state-of-the-art results on two different protein benchmarks of varying complexity. Moreover, the variational design with explicit prior and likelihood functions offers a flexible plug-and-play framework that can be easily customized to suit various protein design tasks.
Authors: Sepehr Mousavi, Shizheng Wen, Levi Lingsch, Maximilian Herde, Bogdan Raoni\'c, Siddhartha Mishra
Abstract: Learning the solution operators of PDEs on arbitrary domains is challenging due to the diversity of possible domain shapes, in addition to the often intricate underlying physics. We propose an end-to-end graph neural network (GNN) based neural operator to learn PDE solution operators from data on point clouds in arbitrary domains. Our multi-scale model maps data between input/output point clouds by passing it through a downsampled regional mesh. Many novel elements are also incorporated to ensure resolution invariance and temporal continuity. Our model, termed RIGNO, is tested on a challenging suite of benchmarks, composed of various time-dependent and steady PDEs defined on a diverse set of domains. We demonstrate that RIGNO is significantly more accurate than neural operator baselines and robustly generalizes to unseen spatial resolutions and time instances.
Authors: Alexander Kozachinskiy, Felipe Urrutia, Hector Jimenez, Tomasz Steifer, Germ\'an Pizarro, Mat\'ias Fuentes, Francisco Meza, Cristian Buc, Crist\'obal Rojas
Abstract: We propose a novel method to evaluate the theoretical limits of Transformers, allowing us to prove the first lower bounds against one-layer softmax Transformers with infinite precision. We establish those bounds for three tasks that require advanced reasoning. The first task, Match3 (Sanford et al., 2023), requires looking at all triples of positions. The second and third tasks address compositionality-based reasoning: one is composition of functions (Peng et al., 2024) and the other is composition of binary relations. We formally prove the inability of one-layer softmax Transformers to solve any of these tasks. In an attempt to overcome these limitations, we introduce Strassen attention and prove that with this mechanism a one-layer Transformer can in principle solve all these tasks. We also show that it enjoys sub-cubic running-time complexity, making it more scalable than similar previously proposed mechanisms, such as higher-order attention (Sanford et al., 2023). To complement our theoretical findings, we experimentally studied Strassen attention and compared it against standard (Vaswani et al, 2017), higher-order attention (Sanford et al., 2023) and triangular attention (Bergen et al. 2021). Our results help to disentangle all these attention mechanisms, highlighting their strengths and limitations. In particular, Strassen attention outperforms standard attention significantly on all the tasks. Altogether, understanding the theoretical limitations can guide research towards scalable attention mechanisms that improve the reasoning abilities of Transformers.
Authors: Yunyang Li, Lin Huang, Zhihao Ding, Chu Wang, Xinran Wei, Han Yang, Zun Wang, Chang Liu, Yu Shi, Peiran Jin, Jia Zhang, Mark Gerstein, Tao Qin
Abstract: Equivariant Graph Neural Networks (EGNNs) have demonstrated significant success in modeling microscale systems, including those in chemistry, biology and materials science. However, EGNNs face substantial computational challenges due to the high cost of constructing edge features via spherical tensor products, making them impractical for large-scale systems. To address this limitation, we introduce E2Former, an equivariant and efficient transformer architecture that incorporates the Wigner $6j$ convolution (Wigner $6j$ Conv). By shifting the computational burden from edges to nodes, the Wigner $6j$ Conv reduces the complexity from $O(|\mathcal{E}|)$ to $ O(| \mathcal{V}|)$ while preserving both the model's expressive power and rotational equivariance. We show that this approach achieves a 7x-30x speedup compared to conventional $\mathrm{SO}(3)$ convolutions. Furthermore, our empirical results demonstrate that the derived E2Former mitigates the computational challenges of existing approaches without compromising the ability to capture detailed geometric information. This development could suggest a promising direction for scalable and efficient molecular modeling.
Authors: Aleksei Kychkin, Georgios C. Chasparis
Abstract: Electricity load consumption may be extremely complex in terms of profile patterns, as it depends on a wide range of human factors, and it is often correlated with several exogenous factors, such as the availability of renewable energy and the weather conditions. The first goal of this paper is to investigate the performance of a large selection of different types of forecasting models in predicting the electricity load consumption within the short time horizon of a day or few hours ahead. Such forecasts may be rather useful for the energy management of individual residential buildings or small energy communities. In particular, we introduce persistence models, standard auto-regressive-based machine learning models, and more advanced deep learning models. The second goal of this paper is to introduce two alternative modeling approaches that are simpler in structure while they take into account domain specific knowledge, as compared to the previously mentioned black-box modeling techniques. In particular, we consider the persistence-based auto-regressive model (PAR) and the seasonal persistence-based regressive model (SPR), priorly introduced by the authors. In this paper, we specifically tailor these models to accommodate the generation of hourly forecasts. The introduced models and the induced comparative analysis extend prior work of the authors which was restricted to day-ahead forecasts. We observed a 15-30% increase in the prediction accuracy of the newly introduced hourly-based forecasting models over existing approaches.
Authors: Xingyu Wang, Mengfan Xu
Abstract: We study decentralized multi-agent multi-armed bandits in fully heavy-tailed settings, where clients communicate over sparse random graphs with heavy-tailed degree distributions and observe heavy-tailed (homogeneous or heterogeneous) reward distributions with potentially infinite variance. The objective is to maximize system performance by pulling the globally optimal arm with the highest global reward mean across all clients. We are the first to address such fully heavy-tailed scenarios, which capture the dynamics and challenges in communication and inference among multiple clients in real-world systems. In homogeneous settings, our algorithmic framework exploits hub-like structures unique to heavy-tailed graphs, allowing clients to aggregate rewards and reduce noises via hub estimators when constructing UCB indices; under $M$ clients and degree distributions with power-law index $\alpha > 1$, our algorithm attains a regret bound (almost) of order $O(M^{1 -\frac{1}{\alpha}} \log{T})$. Under heterogeneous rewards, clients synchronize by communicating with neighbors, aggregating exchanged estimators in UCB indices; With our newly established information delay bounds on sparse random graphs, we prove a regret bound of $O(M \log{T})$. Our results improve upon existing work, which only address time-invariant connected graphs, or light-tailed dynamics in dense graphs and rewards.
Authors: Vladimir Ja\'cimovi\'c, Aladin Crnki\'c
Abstract: The idea of representations of the data in negatively curved manifolds recently attracted a lot of attention and gave a rise to the new research direction named {\it hyperbolic machine learning} (ML). In order to unveil the full potential of this new paradigm, efficient techniques for data analysis and statistical modeling in hyperbolic spaces are necessary. In the present paper rigorous mathematical framework for clustering in hyperbolic spaces is established. First, we introduce the $k$-means clustering in hyperbolic balls, based on the novel definition of barycenter. Second, we present the expectation-maximization (EM) algorithm for learning mixtures of novel probability distributions in hyperbolic balls. In such a way we lay the foundation of unsupervised learning in hyperbolic spaces.
Authors: Xinyu Liu, Zixuan Xie, Shangtong Zhang
Abstract: $Q$-learning is one of the most fundamental reinforcement learning algorithms. Previously, it is widely believed that $Q$-learning with linear function approximation (i.e., linear $Q$-learning) suffers from possible divergence. This paper instead establishes the first $L^2$ convergence rate of linear $Q$-learning to a bounded set. Notably, we do not make any modification to the original linear $Q$-learning algorithm, do not make any Bellman completeness assumption, and do not make any near-optimality assumption on the behavior policy. All we need is an $\epsilon$-softmax behavior policy with an adaptive temperature. The key to our analysis is the general result of stochastic approximations under Markovian noise with fast-changing transition functions. As a side product, we also use this general result to establish the $L^2$ convergence rate of tabular $Q$-learning with an $\epsilon$-softmax behavior policy, for which we rely on a novel pseudo-contraction property of the weighted Bellman optimality operator.
Authors: Pedro Miguel S\'anchez S\'anchez, Enrique Tom\'as Mart\'inez Beltr\'an, Chao Feng, G\'er\^ome Bovet, Gregorio Mart\'inez P\'erez, Alberto Huertas Celdr\'an
Abstract: Decentralized Federated Learning (DFL) enables collaborative, privacy-preserving model training without relying on a central server. This decentralized approach reduces bottlenecks and eliminates single points of failure, enhancing scalability and resilience. However, DFL also introduces challenges such as suboptimal models with non-IID data distributions, increased communication overhead, and resource usage. Thus, this work proposes S-VOTE, a voting-based client selection mechanism that optimizes resource usage and enhances model performance in federations with non-IID data conditions. S-VOTE considers an adaptive strategy for spontaneous local training that addresses participation imbalance, allowing underutilized clients to contribute without significantly increasing resource costs. Extensive experiments on benchmark datasets demonstrate the S-VOTE effectiveness. More in detail, it achieves lower communication costs by up to 21%, 4-6% faster convergence, and improves local performance by 9-17% compared to baseline methods in some configurations, all while achieving a 14-24% energy consumption reduction. These results highlight the potential of S-VOTE to address DFL challenges in heterogeneous environments.
Authors: Antoine de Mathelin, Nicolas Enrique Cecchi, Fran\c{c}ois Deheeger, Mathilde Mougeot, Nicolas Vayatis
Abstract: This paper proposes a novel k-medoids approximation algorithm to handle large-scale datasets with reasonable computational time and memory complexity. We develop a local-search algorithm that iteratively improves the medoid selection based on the estimation of the k-medoids objective. A single batch of size m << n provides the estimation, which reduces the required memory size and the number of pairwise dissimilarities computations to O(mn), instead of O(n^2) compared to most k-medoids baselines. We obtain theoretical results highlighting that a batch of size m = O(log(n)) is sufficient to guarantee, with strong probability, the same performance as the original local-search algorithm. Multiple experiments conducted on real datasets of various sizes and dimensions show that our algorithm provides similar performances as state-of-the-art methods such as FasterPAM and BanditPAM++ with a drastically reduced running time.
Authors: James Flemings, Haosheng Gan, Hongyi Li, Meisam Razaviyayn, Murali Annavaram
Abstract: In-context learning (ICL) has shown promising improvement in downstream task adaptation of LLMs by augmenting prompts with relevant input-output examples (demonstrations). However, the ICL demonstrations can contain privacy-sensitive information, which can be leaked and/or regurgitated by the LLM output. Differential Privacy (DP), a widely adopted privacy safeguard, has emerged to mitigate this privacy leakage, with recent work demonstrating strong privacy-utility tradeoffs in classification tasks for ICL. However, generation tasks for ICL are challenging due to the high-dimensional output space of open-ended generation. To this end, we propose $\texttt{dps-mozo}$, Differentially Private Sampling by Mixing One-shot with Zero-shot Outputs, a decoding framework that generates DP text by sampling from the product of multiple one-shot outputs mixed with a zero-shot output. This mixing effectively reduces the amount of information that can be leaked by each demonstration. By utilizing the inherent randomness in sampling from the mixed distributions, we can achieve DP without adding noise, thereby improving the privacy-utility tradeoff. Our experimental evaluations show $\texttt{dps-mozo}$ can achieve a strong privacy guarantee, $\epsilon=2$, with minimal utility degradation compared to non-private few-shot learning, $\textbf{0.3}$% ROUGE-L F1 score decrease on the SAMSum dataset with Gemma 2 2B.
Authors: Xutong Liu, Xiangxiang Dai, Jinhang Zuo, Siwei Wang, Carlee-Joe Wong, John C. S. Lui, Wei Chen
Abstract: The combinatorial multi-armed bandit (CMAB) is a fundamental sequential decision-making framework, extensively studied over the past decade. However, existing work primarily focuses on the online setting, overlooking the substantial costs of online interactions and the readily available offline datasets. To overcome these limitations, we introduce Off-CMAB, the first offline learning framework for CMAB. Central to our framework is the combinatorial lower confidence bound (CLCB) algorithm, which combines pessimistic reward estimations with combinatorial solvers. To characterize the quality of offline datasets, we propose two novel data coverage conditions and prove that, under these conditions, CLCB achieves a near-optimal suboptimality gap, matching the theoretical lower bound up to a logarithmic factor. We validate Off-CMAB through practical applications, including learning to rank, large language model (LLM) caching, and social influence maximization, showing its ability to handle nonlinear reward functions, general feedback models, and out-of-distribution action samples that excludes optimal or even feasible actions. Extensive experiments on synthetic and real-world datasets further highlight the superior performance of CLCB.
Authors: Gregor Bachmann, Sotiris Anagnostidis, Albert Pumarola, Markos Georgopoulos, Artsiom Sanakoyeu, Yuming Du, Edgar Sch\"onfeld, Ali Thabet, Jonas Kohler
Abstract: The performance of large language models (LLMs) is closely linked to their underlying size, leading to ever-growing networks and hence slower inference. Speculative decoding has been proposed as a technique to accelerate autoregressive generation, leveraging a fast draft model to propose candidate tokens, which are then verified in parallel based on their likelihood under the target model. While this approach guarantees to reproduce the target output, it incurs a substantial penalty: many high-quality draft tokens are rejected, even when they represent objectively valid continuations. Indeed, we show that even powerful draft models such as GPT-4o, as well as human text cannot achieve high acceptance rates under the standard verification scheme. This severely limits the speedup potential of current speculative decoding methods, as an early rejection becomes overwhelmingly likely when solely relying on alignment of draft and target. We thus ask the following question: Can we adapt verification to recognize correct, but non-aligned replies? To this end, we draw inspiration from the LLM-as-a-judge framework, which demonstrated that LLMs are able to rate answers in a versatile way. We carefully design a dataset to elicit the same capability in the target model by training a compact module on top of the embeddings to produce ``judgements" of the current continuation. We showcase our strategy on the Llama-3.1 family, where our 8b/405B-Judge achieves a speedup of 9x over Llama-405B, while maintaining its quality on a large range of benchmarks. These benefits remain present even in optimized inference frameworks, where our method reaches up to 141 tokens/s for 8B/70B-Judge and 129 tokens/s for 8B/405B on 2 and 8 H100s respectively.
Authors: Jan Pauls, Max Zimmer, Berkant Turan, Sassan Saatchi, Philippe Ciais, Sebastian Pokutta, Fabian Gieseke
Abstract: With the rise in global greenhouse gas emissions, accurate large-scale tree canopy height maps are essential for understanding forest structure, estimating above-ground biomass, and monitoring ecological disruptions. To this end, we present a novel approach to generate large-scale, high-resolution canopy height maps over time. Our model accurately predicts canopy height over multiple years given Sentinel-2 time series satellite data. Using GEDI LiDAR data as the ground truth for training the model, we present the first 10m resolution temporal canopy height map of the European continent for the period 2019-2022. As part of this product, we also offer a detailed canopy height map for 2020, providing more precise estimates than previous studies. Our pipeline and the resulting temporal height map are publicly available, enabling comprehensive large-scale monitoring of forests and, hence, facilitating future research and ecological analyses. For an interactive viewer, see https://europetreemap.projects.earthengine.app/view/temporalcanopyheight.
URLs: https://europetreemap.projects.earthengine.app/view/temporalcanopyheight.
Authors: Natalie Maus, Kyurae Kim, Yimeng Zeng, Haydn Thomas Jones, Fangping Wan, Marcelo Der Torossian Torres, Cesar de la Fuente-Nunez, Jacob R. Gardner
Abstract: In multi-objective black-box optimization, the goal is typically to find solutions that optimize a set of T black-box objective functions, $f_1$, ..., $f_T$, simultaneously. Traditional approaches often seek a single Pareto-optimal set that balances trade-offs among all objectives. In this work, we introduce a novel problem setting that departs from this paradigm: finding a smaller set of K solutions, where K < T, that collectively "covers" the T objectives. A set of solutions is defined as "covering" if, for each objective $f_1$, ..., $f_T$, there is at least one good solution. A motivating example for this problem setting occurs in drug design. For example, we may have T pathogens and aim to identify a set of K < T antibiotics such that at least one antibiotic can be used to treat each pathogen. To address this problem, we propose Multi-Objective Coverage Bayesian Optimization (MOCOBO), a principled algorithm designed to efficiently find a covering set. We validate our approach through extensive experiments on challenging high-dimensional tasks, including applications in peptide and molecular design. Experiments demonstrate MOCOBO's ability to find high-performing covering sets of solutions. Additionally, we show that the small sets of K < T solutions found by MOCOBO can match or nearly match the performance of T individually optimized solutions for the same objectives. Our results highlight MOCOBO's potential to tackle complex multi-objective problems in domains where finding at least one high-performing solution for each objective is critical.
Authors: Masahiro Kato, Fumiaki Kozai, Ryo Inokuchi
Abstract: The estimation of average treatment effects (ATEs), defined as the difference in expected outcomes between treatment and control groups, is a central topic in causal inference. This study develops semiparametric efficient estimators for ATE estimation in a setting where only a treatment group and an unknown group-comprising units for which it is unclear whether they received the treatment or control-are observable. This scenario represents a variant of learning from positive and unlabeled data (PU learning) and can be regarded as a special case of ATE estimation with missing data. For this setting, we derive semiparametric efficiency bounds, which provide lower bounds on the asymptotic variance of regular estimators. We then propose semiparametric efficient ATE estimators whose asymptotic variance aligns with these efficiency bounds. Our findings contribute to causal inference with missing data and weakly supervised learning.
Authors: Svein Anders Tunheim, Yujin Zheng, Lei Jiao, Rishad Shafik, Alex Yakovlev, Ole-Christoffer Granmo
Abstract: We present an all-digital programmable machine learning accelerator chip for image classification, underpinning on the Tsetlin machine (TM) principles. The TM is a machine learning algorithm founded on propositional logic, utilizing sub-pattern recognition expressions called clauses. The accelerator implements the coalesced TM version with convolution, and classifies booleanized images of 28$\times$28 pixels with 10 categories. A configuration with 128 clauses is used in a highly parallel architecture. Fast clause evaluation is obtained by keeping all clause weights and Tsetlin automata (TA) action signals in registers. The chip is implemented in a 65 nm low-leakage CMOS technology, and occupies an active area of 2.7mm$^2$. At a clock frequency of 27.8 MHz, the accelerator achieves 60.3k classifications per second, and consumes 8.6 nJ per classification. The latency for classifying a single image is 25.4 $\mu$s which includes system timing overhead. The accelerator achieves 97.42%, 84.54% and 82.55% test accuracies for the datasets MNIST, Fashion-MNIST and Kuzushiji-MNIST, respectively, matching the TM software models.
Authors: Yesom Park, Stanley Osher
Abstract: This paper presents an implicit solution formula for the Hamilton-Jacobi partial differential equation (HJ PDE). The formula is derived using the method of characteristics and is shown to coincide with the Hopf and Lax formulas in the case where either the Hamiltonian or the initial function is convex. It provides a simple and efficient numerical approach for computing the viscosity solution of HJ PDEs, bypassing the need for the Legendre transform of the Hamiltonian or the initial condition, and the explicit computation of individual characteristic trajectories. A deep learning-based methodology is proposed to learn this implicit solution formula, leveraging the mesh-free nature of deep learning to ensure scalability for high-dimensional problems. Building upon this framework, an algorithm is developed that approximates the characteristic curves piecewise linearly for state-dependent Hamiltonians. Extensive experimental results demonstrate that the proposed method delivers highly accurate solutions, even for nonconvex Hamiltonians, and exhibits remarkable scalability, achieving computational efficiency for problems up to 40 dimensions.
Authors: Yuchun Miao, Sen Zhang, Liang Ding, Yuqi Zhang, Lefei Zhang, Dacheng Tao
Abstract: This work identifies the Energy Loss Phenomenon in Reinforcement Learning from Human Feedback (RLHF) and its connection to reward hacking. Specifically, energy loss in the final layer of a Large Language Model (LLM) gradually increases during the RL process, with an excessive increase in energy loss characterizing reward hacking. Beyond empirical analysis, we further provide a theoretical foundation by proving that, under mild conditions, the increased energy loss reduces the upper bound of contextual relevance in LLMs, which is a critical aspect of reward hacking as the reduced contextual relevance typically indicates overfitting to reward model-favored patterns in RL. To address this issue, we propose an Energy loss-aware PPO algorithm (EPPO) which penalizes the increase in energy loss in the LLM's final layer during reward calculation to prevent excessive energy loss, thereby mitigating reward hacking. We theoretically show that EPPO can be conceptually interpreted as an entropy-regularized RL algorithm, which provides deeper insights into its effectiveness. Extensive experiments across various LLMs and tasks demonstrate the commonality of the energy loss phenomenon, as well as the effectiveness of \texttt{EPPO} in mitigating reward hacking and improving RLHF performance.
Authors: Javier Sol\'is-Garc\'ia, Bel\'en Vega-M\'arquez, Juan A. Nepomuceno, Isabel A. Nepomuceno-Chamorro
Abstract: Multivariate Time Series Imputation (MTSI) is crucial for many applications, such as healthcare monitoring and traffic management, where incomplete data can compromise decision-making. Existing state-of-the-art methods, like Denoising Diffusion Probabilistic Models (DDPMs), achieve high imputation accuracy; however, they suffer from significant computational costs and are notably time-consuming due to their iterative nature. In this work, we propose CoSTI, an innovative adaptation of Consistency Models (CMs) for the MTSI domain. CoSTI employs Consistency Training to achieve comparable imputation quality to DDPMs while drastically reducing inference times, making it more suitable for real-time applications. We evaluate CoSTI across multiple datasets and missing data scenarios, demonstrating up to a 98% reduction in imputation time with performance on par with diffusion-based models. This work bridges the gap between efficiency and accuracy in generative imputation tasks, providing a scalable solution for handling missing data in critical spatio-temporal systems.
Authors: Christopher Subich, Syed Zahid Husain, Leo Separovic, Jing Yang
Abstract: Recent advancements in data-driven weather forecasting models have delivered deterministic models that outperform the leading operational forecast systems based on traditional, physics-based models. However, these data-driven models are typically trained with a mean squared error loss function, which causes smoothing of fine scales through a "double penalty" effect. We develop a simple, parameter-free modification to this loss function that avoids this problem by separating the loss attributable to decorrelation from the loss attributable to spectral amplitude errors. Fine-tuning the GraphCast model with this new loss function results in sharp deterministic weather forecasts, an increase of the model's effective resolution from 1,250km to 160km, improvements to ensemble spread, and improvements to predictions of tropical cyclone strength and surface wind extremes.
Authors: Xingyou Song, Dara Bahri
Abstract: Language models have recently been shown capable of performing regression tasks wherein numeric predictions are represented as decoded strings. In this work, we provide theoretical grounds for this capability and furthermore investigate the utility of causal auto-regressive sequence models when they are applied to any feature representation. We find that, despite being trained in the usual way - for next-token prediction via cross-entropy loss - decoding-based regression is as performant as traditional approaches for tabular regression tasks, while being flexible enough to capture arbitrary distributions, such as in the task of density estimation.
Authors: Wenzhi Fang, Dong-Jun Han, Liangqi Yuan, Seyyedali Hosseinalipour, Christopher G. Brinton
Abstract: Fine-tuning large language models (LLMs) on devices is attracting increasing interest. Recent works have fused low-rank adaptation (LoRA) techniques with federated fine-tuning to mitigate challenges associated with device model sizes and data scarcity. Still, the heterogeneity of computational resources remains a critical bottleneck: while higher-rank modules generally enhance performance, varying device capabilities constrain LoRA's feasible rank range. Existing approaches attempting to resolve this issue either lack analytical justification or impose additional computational overhead, leaving a wide gap for an efficient and theoretically-grounded solution. To address these challenges, we propose federated sketching LoRA (FSLoRA), which leverages a sketching mechanism to enable devices to selectively update submatrices of global LoRA modules maintained by the server. By adjusting the sketching ratios, which determine the ranks of the submatrices on the devices, FSLoRA flexibly adapts to device-specific communication and computational constraints. We provide a rigorous convergence analysis of FSLoRA that characterizes how the sketching ratios affect the convergence rate. Through comprehensive experiments on multiple datasets and LLM models, we demonstrate FSLoRA's superior performance compared to various baselines.
Authors: Alina Shutova, Vladimir Malinovskii, Vage Egiazarian, Denis Kuznedelev, Denis Mazur, Nikita Surkov, Ivan Ermakov, Dan Alistarh
Abstract: Efficient real-world deployments of large language models (LLMs) rely on Key-Value (KV) caching for processing and generating long outputs, reducing the need for repetitive computation. For large contexts, Key-Value caches can take up tens of gigabytes of device memory, as they store vector representations for each token and layer. Recent work has shown that the cached vectors can be compressed through quantization, pruning or merging, but these techniques often compromise quality towards higher compression rates. In this work, we aim to improve Key & Value compression by exploiting two observations: 1) the inherent dependencies between keys and values across different layers, and 2) high-compression mechanisms for internal network states. We propose AQUA-KV, an adaptive quantization for Key-Value caches that relies on compact adapters to exploit existing dependencies between Keys and Values, and aims to "optimally" compress the information that cannot be predicted. AQUA-KV significantly improves compression rates, while maintaining high accuracy on state-of-the-art LLM families. On Llama 3.2 LLMs, we achieve near-lossless inference at 2-2.5 bits per value with under $1\%$ relative error in perplexity and LongBench scores. AQUA-KV is one-shot, simple, and efficient: it can be calibrated on a single GPU within 1-6 hours, even for 70B models.
Authors: Andrey Polubarov, Nikita Lyubaykin, Alexander Derevyagin, Ilya Zisman, Denis Tarasov, Alexander Nikulin, Vladislav Kurenkov
Abstract: In-Context Reinforcement Learning (ICRL) represents a promising paradigm for developing generalist agents that learn at inference time through trial-and-error interactions, analogous to how large language models adapt contextually, but with a focus on reward maximization. However, the scalability of ICRL beyond toy tasks and single-domain settings remains an open challenge. In this work, we present the first steps toward scaling ICRL by introducing a fixed, cross-domain model capable of learning behaviors through in-context reinforcement learning. Our results demonstrate that Algorithm Distillation, a framework designed to facilitate ICRL, offers a compelling and competitive alternative to expert distillation to construct versatile action models. These findings highlight the potential of ICRL as a scalable approach for generalist decision-making systems. Code to be released at https://github.com/dunnolab/vintix
Authors: Argyrios Gerogiannis, Yu-Han Huang, Subhonmesh Bose, Venugopal V. Veeravalli
Abstract: We study the problem of piecewise stationary bandits without prior knowledge of the underlying non-stationarity. We propose the first $\textit{feasible}$ black-box algorithm applicable to most common parametric bandit variants. Our procedure, termed Detection Augmented Bandit (DAB), is modular, accepting any stationary bandit algorithm as input and augmenting it with a change detector. DAB achieves optimal regret in the piecewise stationary setting under mild assumptions. Specifically, we prove that DAB attains the order-optimal regret bound of $\tilde{\mathcal{O}}(\sqrt{N_T T})$, where $N_T$ denotes the number of changes over the horizon $T$, if its input stationary bandit algorithm has order-optimal stationary regret guarantees. Applying DAB to different parametric bandit settings, we recover recent state-of-the-art results. Notably, for self-concordant bandits, DAB achieves optimal dynamic regret, while previous works obtain suboptimal bounds and require knowledge on the non-stationarity. In simulations on piecewise stationary environments, DAB outperforms existing approaches across varying number of changes. Interestingly, despite being theoretically designed for piecewise stationary environments, DAB is also effective in simulations in drifting environments, outperforming existing methods designed specifically for this scenario.
Authors: Yingdan Shi, Ren Wang
Abstract: Machine unlearning seeks to systematically remove specified data from a trained model, effectively achieving a state as though the data had never been encountered during training. While metrics such as Unlearning Accuracy (UA) and Membership Inference Attack (MIA) provide a baseline for assessing unlearning performance, they fall short of evaluating the completeness and reliability of forgetting. This is because the ground truth labels remain potential candidates within the scope of uncertainty quantification, leaving gaps in the evaluation of true forgetting. In this paper, we identify critical limitations in existing unlearning metrics and propose enhanced evaluation metrics inspired by conformal prediction. Our metrics can effectively capture the extent to which ground truth labels are excluded from the prediction set. Furthermore, we observe that many existing machine unlearning methods do not achieve satisfactory forgetting performance when evaluated with our new metrics. To address this, we propose an unlearning framework that integrates conformal prediction insights into Carlini & Wagner adversarial attack loss. Extensive experiments on the image classification task demonstrate that our enhanced metrics offer deeper insights into unlearning effectiveness, and that our unlearning framework significantly improves the forgetting quality of unlearning methods.
Authors: Matthew Chen, Joshua Engels, Max Tegmark
Abstract: Sparse autoencoders (SAEs) decompose language model representations into a sparse set of linear latent vectors. Recent works have improved SAEs using language model gradients, but these techniques require many expensive backward passes during training and still cause a significant increase in cross entropy loss when SAE reconstructions are inserted into the model. In this work, we improve on these limitations by taking a fundamentally different approach: we use low-rank adaptation (LoRA) to finetune the language model itself around a previously trained SAE. We analyze our method across SAE sparsity, SAE width, language model size, LoRA rank, and model layer on the Gemma Scope family of SAEs. In these settings, our method reduces the cross entropy loss gap by 30% to 55% when SAEs are inserted during the forward pass. We also find that compared to end-to-end (e2e) SAEs, our approach achieves the same downstream cross entropy loss 3$\times$ to 20$\times$ faster on Gemma-2-2B and 2$\times$ to 10$\times$ faster on Llama-3.2-1B. We further show that our technique improves downstream metrics and can adapt multiple SAEs at once. Our results demonstrate that improving model interpretability is not limited to post-hoc SAE training; Pareto improvements can also be achieved by directly optimizing the model itself.
Authors: Zhen Guo, Reza Tourani
Abstract: With the growing demand for personalized AI solutions, customized LLMs have become a preferred choice for businesses and individuals, driving the deployment of millions of AI agents across various platforms, e.g., GPT Store hosts over 3 million customized GPTs. Their popularity is partly driven by advanced reasoning capabilities, such as Chain-of-Thought, which enhance their ability to tackle complex tasks. However, their rapid proliferation introduces new vulnerabilities, particularly in reasoning processes that remain largely unexplored. We introduce DarkMind, a novel backdoor attack that exploits the reasoning capabilities of customized LLMs. Designed to remain latent, DarkMind activates within the reasoning chain to covertly alter the final outcome. Unlike existing attacks, it operates without injecting triggers into user queries, making it a more potent threat. We evaluate DarkMind across eight datasets covering arithmetic, commonsense, and symbolic reasoning domains, using five state-of-the-art LLMs with five distinct trigger implementations. Our results demonstrate DarkMind effectiveness across all scenarios, underscoring its impact. Finally, we explore potential defense mechanisms to mitigate its risks, emphasizing the need for stronger security measures.
Authors: Yuexing Han, Ruijie Li
Abstract: Deep learning models have been widely applied across various domains and industries. However, many fields still face challenges due to limited and insufficient data. This paper proposes a Feature Augmentation on Adaptive Geodesic Curve (FAAGC) method in the pre-shape space to increase data. In the pre-shape space, objects with identical shapes lie on a great circle. Thus, we project deep model representations into the pre-shape space and construct a geodesic curve, i.e., an arc of a great circle, for each class. Feature augmentation is then performed by sampling along these geodesic paths. Extensive experiments demonstrate that FAAGC improves classification accuracy under data-scarce conditions and generalizes well across various feature types.
Authors: Gerrit Klause, Niklas Bunzel
Abstract: Neural networks are vulnerable to adversarial attacks, and several defenses have been proposed. Designing a robust network is a challenging task given the wide range of attacks that have been developed. Therefore, we aim to provide insight into the influence of network similarity on the success rate of transferred adversarial attacks. Network designers can then compare their new network with existing ones to estimate its vulnerability. To achieve this, we investigate the complex relationship between network similarity and the success rate of transferred adversarial attacks. We applied the Centered Kernel Alignment (CKA) network similarity score and used various methods to find a correlation between a large number of Convolutional Neural Networks (CNNs) and adversarial attacks. Network similarity was found to be moderate across different CNN architectures, with more complex models such as DenseNet showing lower similarity scores due to their architectural complexity. Layer similarity was highest for consistent, basic layers such as DataParallel, Dropout and Conv2d, while specialized layers showed greater variability. Adversarial attack success rates were generally consistent for non-transferred attacks, but varied significantly for some transferred attacks, with complex networks being more vulnerable. We found that a DecisionTreeRegressor can predict the success rate of transferred attacks for all black-box and Carlini & Wagner attacks with an accuracy of over 90%, suggesting that predictive models may be viable under certain conditions. However, the variability of results across different data subsets underscores the complexity of these relationships and suggests that further research is needed to generalize these findings across different attack scenarios and network architectures.
Authors: Sheila E. Whitman, Marat I. Latypov
Abstract: Machine learning of microstructure--property relationships from data is an emerging approach in computational materials science. Most existing machine learning efforts focus on the development of task-specific models for each microstructure--property relationship. We propose utilizing pre-trained foundational vision transformers for the extraction of task-agnostic microstructure features and subsequent light-weight machine learning of a microstructure-dependent property. We demonstrate our approach with pre-trained state-of-the-art vision transformers (CLIP, DINOV2, SAM) in two case studies on machine-learning: (i) elastic modulus of two-phase microstructures based on simulations data; and (ii) Vicker's hardness of Ni-base and Co-base superalloys based on experimental data published in literature. Our results show the potential of foundational vision transformers for robust microstructure representation and efficient machine learning of microstructure--property relationships without the need for expensive task-specific training or fine-tuning of bespoke deep learning models.
Authors: Rupak Kumar Das, Jonathan Dodge
Abstract: With their advanced capabilities, Large Language Models (LLMs) can generate highly convincing and contextually relevant fake news, which can contribute to disseminating misinformation. Though there is much research on fake news detection for human-written text, the field of detecting LLM-generated fake news is still under-explored. This research measures the efficacy of detectors in identifying LLM-paraphrased fake news, in particular, determining whether adding a paraphrase step in the detection pipeline helps or impedes detection. This study contributes: (1) Detectors struggle to detect LLM-paraphrased fake news more than human-written text, (2) We find which models excel at which tasks (evading detection, paraphrasing to evade detection, and paraphrasing for semantic similarity). (3) Via LIME explanations, we discovered a possible reason for detection failures: sentiment shift. (4) We discover a worrisome trend for paraphrase quality measurement: samples that exhibit sentiment shift despite a high BERTSCORE. (5) We provide a pair of datasets augmenting existing datasets with paraphrase outputs and scores. The dataset is available on GitHub
Authors: Sebastian Pena, Lin Lin, Jia Li
Abstract: The rapid emergence of single-cell data has facilitated the study of many different biological conditions at the cellular level. Cluster analysis has been widely applied to identify cell types, capturing the essential patterns of the original data in a much more concise form. One challenge in the cluster analysis of cells is matching clusters extracted from datasets of different origins or conditions. Many existing algorithms cannot recognize new cell types present in only one of the two samples when establishing a correspondence between clusters obtained from two samples. Additionally, when there are more than two samples, it is advantageous to align clusters across all samples simultaneously rather than performing pairwise alignment. Our approach aims to construct a taxonomy for cell clusters across all samples to better annotate these clusters and effectively extract features for downstream analysis. A new system for constructing cell-type taxonomy has been developed by combining the technique of Optimal Transport with Relaxed Marginal Constraints (OT-RMC) and the simultaneous alignment of clusters across multiple samples. OT-RMC allows us to address challenges that arise when the proportions of clusters vary substantially between samples or when some clusters do not appear in all the samples. Experiments on more than twenty datasets demonstrate that the taxonomy constructed by this new system can yield highly accurate annotation of cell types. Additionally, sample-level features extracted based on the taxonomy result in accurate classification of samples.
Authors: Mingyang Li, Michelle Kuchera, Raghuram Ramanujan, Adam Anthony, Curtis Hunt, Yassid Ayyad
Abstract: Modeling detector response is a key challenge in time projection chambers. We cast this problem as an unpaired point cloud translation task, between data collected from simulations and from experimental runs. Effective translation can assist with both noise rejection and the construction of high-fidelity simulators. Building on recent work in diffusion probabilistic models, we present a novel framework for performing this mapping. We demonstrate the success of our approach in both synthetic domains and in data sourced from the Active-Target Time Projection Chamber.
Authors: Alfio Quarteroni, Paola Gervasio, Francesco Regazzoni
Abstract: Scientific Machine Learning (SciML) is a recently emerged research field which combines physics-based and data-driven models for the numerical approximation of differential problems. Physics-based models rely on the physical understanding of the problem at hand, subsequent mathematical formulation, and numerical approximation. Data-driven models instead aim to extract relations between input and output data without arguing any causality principle underlining the available data distribution. In recent years, data-driven models have been rapidly developed and popularized. Such a diffusion has been triggered by a huge availability of data (the so-called big data), an increasingly cheap computing power, and the development of powerful machine learning algorithms. SciML leverages the physical awareness of physics-based models and, at the same time, the efficiency of data-driven algorithms. With SciML, we can inject physics and mathematical knowledge into machine learning algorithms. Yet, we can rely on data-driven algorithms' capability to discover complex and non-linear patterns from data and improve the descriptive capacity of physics-based models. After recalling the mathematical foundations of digital modelling and machine learning algorithms, and presenting the most popular machine learning architectures, we discuss the great potential of a broad variety of SciML strategies in solving complex problems governed by partial differential equations. Finally, we illustrate the successful application of SciML to the simulation of the human cardiac function, a field of significant socio-economic importance that poses numerous challenges on both the mathematical and computational fronts. The corresponding mathematical model is a complex system of non-linear ordinary and partial differential equations describing the electromechanics, valve dynamics, blood circulation, perfusion in the coronary tree, and torso potential. Despite the robustness and accuracy of physics-based models, certain aspects, such as unveiling constitutive laws for cardiac cells and myocardial material properties, as well as devising efficient reduced order models to dominate the extraordinary computational complexity, have been successfully tackled by leveraging data-driven models.
Authors: Andrew M Birnbaum, Adam Buchwald, Peter Turkeltaub, Adam Jacks, Yu Huang, Abhisheck Datta, Lucas C Parra, Lukas A Hirsch
Abstract: The goal of this work was to develop a deep network for whole-head segmentation, including clinical MRIs with abnormal anatomy, and compile the first public benchmark dataset for this purpose. We collected 91 MRIs with volumetric segmentation labels for a diverse set of human subjects (4 normal, 32 traumatic brain injuries, and 57 strokes). These clinical cases are characterized by extended cerebrospinal fluid (CSF) in regions normally containing the brain. Training labels were generated by manually correcting initial automated segmentations for skin/scalp, skull, CSF, gray matter, white matter, air cavity, and extracephalic air. We developed a MultiAxial network consisting of three 2D U-Net models that operate independently in sagittal, axial, and coronal planes and are then combined to produce a single 3D segmentation. The MultiAxial network achieved test-set Dice scores of 0.88 (median plus-minus 0.04). For brain tissue, it significantly outperforms existing brain segmentation methods (MultiAxial: 0.898 plus-minus 0.041, SynthSeg: 0.758 plus-minus 0.054, BrainChop: 0.757 plus-minus 0.125). The MultiAxial network gains in robustness by avoiding the need for coregistration with an atlas. It performed well in regions with abnormal anatomy and on images that have been de-identified. It enables more robust current flow modeling when incorporated into ROAST, a widely-used modeling toolbox for transcranial electric stimulation. We are releasing a state-of-the-art model for whole-head MRI segmentation, along with a dataset of 61 clinical MRIs and training labels, including non-brain structures. Together, the model and data may serve as a benchmark for future efforts.
Authors: Konstantinos Mitsides, Maxence Faldor, Antoine Cully
Abstract: Quality-Diversity optimization comprises a family of evolutionary algorithms aimed at generating a collection of diverse and high-performing solutions. MAP-Elites (ME), a notable example, is used effectively in fields like evolutionary robotics. However, the reliance of ME on random mutations from Genetic Algorithms limits its ability to evolve high-dimensional solutions. Methods proposed to overcome this include using gradient-based operators like policy gradients or natural evolution strategies. While successful at scaling ME for neuroevolution, these methods often suffer from slow training speeds, or difficulties in scaling with massive parallelization due to high computational demands or reliance on centralized actor-critic training. In this work, we introduce a fast, sample-efficient ME based algorithm capable of scaling up with massive parallelization, significantly reducing runtimes without compromising performance. Our method, ASCII-ME, unlike existing policy gradient quality-diversity methods, does not rely on centralized actor-critic training. It performs behavioral variations based on time step performance metrics and maps these variations to solutions using policy gradients. Our experiments show that ASCII-ME can generate a diverse collection of high-performing deep neural network policies in less than 250 seconds on a single GPU. Additionally, it operates on average, five times faster than state-of-the-art algorithms while still maintaining competitive sample efficiency.
Authors: Mohd. Farhan Israk Soumik, W. K. M. Mithsara, Abdur R. Shahid, Ahmed Imteaj
Abstract: The rapid proliferation of speech-enabled technologies, including virtual assistants, video conferencing platforms, and wearable devices, has raised significant privacy concerns, particularly regarding the inference of sensitive emotional information from audio data. Existing privacy-preserving methods often compromise usability and security, limiting their adoption in practical scenarios. This paper introduces a novel, user-centric approach that leverages familiar audio editing techniques, specifically pitch and tempo manipulation, to protect emotional privacy without sacrificing usability. By analyzing popular audio editing applications on Android and iOS platforms, we identified these features as both widely available and usable. We rigorously evaluated their effectiveness against a threat model, considering adversarial attacks from diverse sources, including Deep Neural Networks (DNNs), Large Language Models (LLMs), and and reversibility testing. Our experiments, conducted on three distinct datasets, demonstrate that pitch and tempo manipulation effectively obfuscates emotional data. Additionally, we explore the design principles for lightweight, on-device implementation to ensure broad applicability across various devices and platforms.
Authors: Anthony Mendil, Felix Putze
Abstract: Attribute manipulation deals with the problem of changing individual attributes of a data point or a time series, while leaving all other aspects unaffected. This work focuses on the domain of human motion, more precisely karate movement patterns. To the best of our knowledge, it presents the first success at manipulating attributes of human motion data. One of the key requirements for achieving attribute manipulation on human motion is a suitable pose representation. Therefore, we design a novel rotation-based pose representation that enables the disentanglement of the human skeleton and the motion trajectory, while still allowing an accurate reconstruction of the original anatomy. The core idea of the manipulation approach is to use a transformer encoder for discovering high-level semantics, and a diffusion probabilistic model for modeling the remaining stochastic variations. We show that the embedding space obtained from the transformer encoder is semantically meaningful and linear. This enables the manipulation of high-level attributes, by discovering their linear direction of change in the semantic embedding space and moving the embedding along said direction. The code and data are available at https://github.com/anthony-mendil/MoDiffAE.
Authors: Majid Dashtbani, Ladan Tahvildari
Abstract: While cloud environments and auto-scaling solutions have been widely applied to traditional monolithic applications, they face significant limitations when it comes to microservices-based architectures. Microservices introduce additional challenges due to their dynamic and spatiotemporal characteristics, which require more efficient and specialized auto-scaling strategies. Centralized auto-scaling for the entire microservice application is insufficient, as each service within a chain has distinct specifications and performance requirements. Therefore, each service requires its own dedicated auto-scaler to address its unique scaling needs effectively, while also considering the dependencies with other services in the chain and the overall application. This paper presents a combination of control theory, machine learning, and heuristics to address these challenges. We propose an adaptive auto-scaling framework, STaleX, for microservices that integrates spatiotemporal features, enabling real-time resource adjustments to minimize SLO violations. STaleX employs a set of weighted Proportional-Integral-Derivative (PID) controllers for each service, where weights are dynamically adjusted based on a supervisory unit that integrates spatiotemporal features. This supervisory unit continuously monitors and adjusts both the weights and the resources allocated to each service. Our framework accounts for spatial features, including service specifications and dependencies among services, as well as temporal variations in workload, ensuring that resource allocation is continuously optimized. Through experiments on a microservice-based demo application deployed on a Kubernetes cluster, we demonstrate the effectiveness of our framework in improving performance and reducing costs compared to traditional scaling methods like Kubernetes Horizontal Pod Autoscaler (HPA) with a 26.9% reduction in resource usage.
Authors: Nuojin Cheng, Leonard Papenmeier, Stephen Becker, Luigi Nardi
Abstract: Bayesian optimization is a widely used method for optimizing expensive black-box functions, with Expected Improvement being one of the most commonly used acquisition functions. In contrast, information-theoretic acquisition functions aim to reduce uncertainty about the function's optimum and are often considered fundamentally distinct from EI. In this work, we challenge this prevailing perspective by introducing a unified theoretical framework, Variational Entropy Search, which reveals that EI and information-theoretic acquisition functions are more closely related than previously recognized. We demonstrate that EI can be interpreted as a variational inference approximation of the popular information-theoretic acquisition function, named Max-value Entropy Search. Building on this insight, we propose VES-Gamma, a novel acquisition function that balances the strengths of EI and MES. Extensive empirical evaluations across both low- and high-dimensional synthetic and real-world benchmarks demonstrate that VES-Gamma is competitive with state-of-the-art acquisition functions and in many cases outperforms EI and MES.
Authors: David Mart\'inez-Rubio, Sebastian Pokutta
Abstract: We introduce novel techniques to enhance Frank-Wolfe algorithms by leveraging function smoothness beyond traditional short steps. Our study focuses on Frank-Wolfe algorithms with step sizes that incorporate primal-dual guarantees, offering practical stopping criteria. We present a new Frank-Wolfe algorithm utilizing an optimistic framework and provide a primal-dual convergence proof. Additionally, we propose a generalized short-step strategy aimed at optimizing a computable primal-dual gap. Interestingly, this new generalized short-step strategy is also applicable to gradient descent algorithms beyond Frank-Wolfe methods. As a byproduct, our work revisits and refines primal-dual techniques for analyzing Frank-Wolfe algorithms, achieving tighter primal-dual convergence rates. Empirical results demonstrate that our optimistic algorithm outperforms existing methods, highlighting its practical advantages.
Authors: Vitor Guizilini, Muhammad Zubair Irshad, Dian Chen, Greg Shakhnarovich, Rares Ambrus
Abstract: Current methods for 3D scene reconstruction from sparse posed images employ intermediate 3D representations such as neural fields, voxel grids, or 3D Gaussians, to achieve multi-view consistent scene appearance and geometry. In this paper we introduce MVGD, a diffusion-based architecture capable of direct pixel-level generation of images and depth maps from novel viewpoints, given an arbitrary number of input views. Our method uses raymap conditioning to both augment visual features with spatial information from different viewpoints, as well as to guide the generation of images and depth maps from novel views. A key aspect of our approach is the multi-task generation of images and depth maps, using learnable task embeddings to guide the diffusion process towards specific modalities. We train this model on a collection of more than 60 million multi-view samples from publicly available datasets, and propose techniques to enable efficient and consistent learning in such diverse conditions. We also propose a novel strategy that enables the efficient training of larger models by incrementally fine-tuning smaller ones, with promising scaling behavior. Through extensive experiments, we report state-of-the-art results in multiple novel view synthesis benchmarks, as well as multi-view stereo and video depth estimation.
Authors: Andrey Borro, Patricia J Riddle, Michael W Barley, Michael J Witbrock
Abstract: While systems designed for solving planning tasks vastly outperform Large Language Models (LLMs) in this domain, they usually discard the rich semantic information embedded within task descriptions. In contrast, LLMs possess parametrised knowledge across a wide range of topics, enabling them to leverage the natural language descriptions of planning tasks in their solutions. However, current research in this direction faces challenges in generating correct and executable plans. Furthermore, these approaches depend on the LLM to output solutions in an intermediate language, which must be translated into the representation language of the planning task. We introduce a novel planning method, which leverages the parametrised knowledge of LLMs by using their output as a heuristic for Hill-Climbing Search. This approach is further enhanced by prompting the LLM to generate a solution estimate to guide the search. Our method outperforms the task success rate of similar systems within a common household environment by 22 percentage points, with consistently executable plans. All actions are encoded in their original representation, demonstrating that strong results can be achieved without an intermediate language, thus eliminating the need for a translation step.
Authors: Antoine Simoulin, Namyong Park, Xiaoyi Liu, Grey Yang
Abstract: Fine-tuning provides an effective means to specialize pre-trained models for various downstream tasks. However, fine-tuning often incurs high memory overhead, especially for large transformer-based models, such as LLMs. While existing methods may reduce certain parts of the memory required for fine-tuning, they still require caching all intermediate activations computed in the forward pass to update weights during the backward pass. In this work, we develop TokenTune, a method to reduce memory usage, specifically the memory to store intermediate activations, in the fine-tuning of transformer-based models. During the backward pass, TokenTune approximates the gradient computation by backpropagating through just a subset of input tokens. Thus, with TokenTune, only a subset of intermediate activations are cached during the forward pass. Also, TokenTune can be easily combined with existing methods like LoRA, further reducing the memory cost. We evaluate our approach on pre-trained transformer models with up to billions of parameters, considering the performance on multiple downstream tasks such as text classification and question answering in a few-shot learning setup. Overall, TokenTune achieves performance on par with full fine-tuning or representative memory-efficient fine-tuning methods, while greatly reducing the memory footprint, especially when combined with other methods with complementary memory reduction mechanisms. We hope that our approach will facilitate the fine-tuning of large transformers, in specializing them for specific domains or co-training them with other neural components from a larger system. Our code is available at https://github.com/facebookresearch/tokentune.
Authors: Mrinank Sharma, Meg Tong, Jesse Mu, Jerry Wei, Jorrit Kruthoff, Scott Goodfriend, Euan Ong, Alwin Peng, Raj Agarwal, Cem Anil, Amanda Askell, Nathan Bailey, Joe Benton, Emma Bluemke, Samuel R. Bowman, Eric Christiansen, Hoagy Cunningham, Andy Dau, Anjali Gopal, Rob Gilson, Logan Graham, Logan Howard, Nimit Kalra, Taesung Lee, Kevin Lin, Peter Lofgren, Francesco Mosconi, Clare O'Hara, Catherine Olsson, Linda Petrini, Samir Rajani, Nikhil Saxena, Alex Silverstein, Tanya Singh, Theodore Sumers, Leonard Tang, Kevin K. Troy, Constantin Weisser, Ruiqi Zhong, Giulio Zhou, Jan Leike, Jared Kaplan, Ethan Perez
Abstract: Large language models (LLMs) are vulnerable to universal jailbreaks-prompting strategies that systematically bypass model safeguards and enable users to carry out harmful processes that require many model interactions, like manufacturing illegal substances at scale. To defend against these attacks, we introduce Constitutional Classifiers: safeguards trained on synthetic data, generated by prompting LLMs with natural language rules (i.e., a constitution) specifying permitted and restricted content. In over 3,000 estimated hours of red teaming, no red teamer found a universal jailbreak that could extract information from an early classifier-guarded LLM at a similar level of detail to an unguarded model across most target queries. On automated evaluations, enhanced classifiers demonstrated robust defense against held-out domain-specific jailbreaks. These classifiers also maintain deployment viability, with an absolute 0.38% increase in production-traffic refusals and a 23.7% inference overhead. Our work demonstrates that defending against universal jailbreaks while maintaining practical deployment viability is tractable.
Authors: Jiaqi Tang, Yuling Yan
Abstract: Score-based generative models, which transform noise into data by learning to reverse a diffusion process, have become a cornerstone of modern generative AI. This paper contributes to establishing theoretical guarantees for the probability flow ODE, a widely used diffusion-based sampler known for its practical efficiency. While a number of prior works address its general convergence theory, it remains unclear whether the probability flow ODE sampler can adapt to the low-dimensional structures commonly present in natural image data. We demonstrate that, with accurate score function estimation, the probability flow ODE sampler achieves a convergence rate of $O(k/T)$ in total variation distance (ignoring logarithmic factors), where $k$ is the intrinsic dimension of the target distribution and $T$ is the number of iterations. This dimension-free convergence rate improves upon existing results that scale with the typically much larger ambient dimension, highlighting the ability of the probability flow ODE sampler to exploit intrinsic low-dimensional structures in the target distribution for faster sampling.
Authors: Zhengqi Gao, Kaiwen Zha, Tianyuan Zhang, Zihui Xue, Duane S. Boning
Abstract: Guidance techniques are simple yet effective for improving conditional generation in diffusion models. Albeit their empirical success, the practical implementation of guidance diverges significantly from its theoretical motivation. In this paper, we reconcile this discrepancy by replacing the scaled marginal distribution target, which we prove theoretically invalid, with a valid scaled joint distribution objective. Additionally, we show that the established guidance implementations are approximations to the intractable optimal solution under no future foresight constraint. Building on these theoretical insights, we propose rectified gradient guidance (REG), a versatile enhancement designed to boost the performance of existing guidance methods. Experiments on 1D and 2D demonstrate that REG provides a better approximation to the optimal solution than prior guidance techniques, validating the proposed theoretical framework. Extensive experiments on class-conditional ImageNet and text-to-image generation tasks show that incorporating REG consistently improves FID and Inception/CLIP scores across various settings compared to its absence.
Authors: Mingzhi Yuan, Zihan Zou, Wei Hu
Abstract: Developing machine learning protocols for molecular simulations requires comprehensive and efficient datasets. Here we introduce the QMe14S dataset, comprising 186,102 small organic molecules featuring 14 elements (H, B, C, N, O, F, Al, Si, P, S, Cl, As, Se, Br) and 47 functional groups. Using density functional theory at the B3LYP/TZVP level, we optimized the geometries and calculated properties including energy, atomic charge, atomic force, dipole moment, quadrupole moment, polarizability, octupole moment, first hyperpolarizability, and Hessian. At the same level, we obtained the harmonic IR, Raman and NMR spectra. Furthermore, we conducted ab initio molecular dynamics simulations to generate dynamic configurations and extract nonequilibrium properties, including energy, forces, and Hessians. By leveraging our E(3)-equivariant message-passing neural network (DetaNet), we demonstrated that models trained on QMe14S outperform those trained on the previously developed QM9S dataset in simulating molecular spectra. The QMe14S dataset thus serves as a comprehensive benchmark for molecular simulations, offering valuable insights into structure-property relationships.
Authors: Jaesin Ahn, Heechul Jung
Abstract: Text-to-image diffusion models show remarkable generation performance following text prompts, but risk generating Not Safe For Work (NSFW) contents from unsafe prompts. Existing approaches, such as prompt filtering or concept unlearning, fail to defend against adversarial attacks while maintaining benign image quality. In this paper, we propose a novel approach called Distorting Embedding Space (DES), a text encoder-based defense mechanism that effectively tackles these issues through innovative embedding space control. DES transforms unsafe embeddings, extracted from a text encoder using unsafe prompts, toward carefully calculated safe embedding regions to prevent unsafe contents generation, while reproducing the original safe embeddings. DES also neutralizes the nudity embedding, extracted using prompt ``nudity", by aligning it with neutral embedding to enhance robustness against adversarial attacks. These methods ensure both robust defense and high-quality image generation. Additionally, DES can be adopted in a plug-and-play manner and requires zero inference overhead, facilitating its deployment. Extensive experiments on diverse attack types, including black-box and white-box scenarios, demonstrate DES's state-of-the-art performance in both defense capability and benign image generation quality. Our model is available at https://github.com/aei13/DES.
Authors: Joshua R. Waite, Md. Zahid Hasan, Qisai Liu, Zhanhong Jiang, Chinmay Hegde, Soumik Sarkar
Abstract: Vision-language model (VLM) fine-tuning for application-specific visual grounding based on natural language instructions has become one of the most popular approaches for learning-enabled autonomous systems. However, such fine-tuning relies heavily on high-quality datasets to achieve successful performance in various downstream tasks. Additionally, VLMs often encounter limitations due to insufficient and imbalanced fine-tuning data. To address these issues, we propose a new generalizable framework to improve VLM fine-tuning by integrating it with a reinforcement learning (RL) agent. Our method utilizes the RL agent to manipulate objects within an indoor setting to create synthetic data for fine-tuning to address certain vulnerabilities of the VLM. Specifically, we use the performance of the VLM to provide feedback to the RL agent to generate informative data that efficiently fine-tune the VLM over the targeted task (e.g. spatial reasoning). The key contribution of this work is developing a framework where the RL agent serves as an informative data sampling tool and assists the VLM in order to enhance performance and address task-specific vulnerabilities. By targeting the data sampling process to address the weaknesses of the VLM, we can effectively train a more context-aware model. In addition, generating synthetic data allows us to have precise control over each scene and generate granular ground truth captions. Our results show that the proposed data generation approach improves the spatial reasoning performance of VLMs, which demonstrates the benefits of using RL-guided data generation in vision-language tasks.
Authors: Jae Yong Lee, Sungmin Kang, Shin Yoo
Abstract: Large Language Models (LLMs) are machine learning models that have seen widespread adoption due to their capability of handling previously difficult tasks. LLMs, due to their training, are sensitive to how exactly a question is presented, also known as prompting. However, prompting well is challenging, as it has been difficult to uncover principles behind prompting -- generally, trial-and-error is the most common way of improving prompts, despite its significant computational cost. In this context, we argue it would be useful to perform `predictive prompt analysis', in which an automated technique would perform a quick analysis of a prompt and predict how the LLM would react to it, relative to a goal provided by the user. As a demonstration of the concept, we present Syntactic Prevalence Analyzer (SPA), a predictive prompt analysis approach based on sparse autoencoders (SAEs). SPA accurately predicted how often an LLM would generate target syntactic structures during code synthesis, with up to 0.994 Pearson correlation between the predicted and actual prevalence of the target structure. At the same time, SPA requires only 0.4\% of the time it takes to run the LLM on a benchmark. As LLMs are increasingly used during and integrated into modern software development, our proposed predictive prompt analysis concept has the potential to significantly ease the use of LLMs for both practitioners and researchers.
Authors: Zijun Gao, Yan Sun
Abstract: Generative AI (GenAI) models have recently achieved remarkable empirical performance in various applications, however, their evaluations yet lack uncertainty quantification. In this paper, we propose a method to compare two generative models based on an unbiased estimator of their relative performance gap. Statistically, our estimator achieves parametric convergence rate and asymptotic normality, which enables valid inference. Computationally, our method is efficient and can be accelerated by parallel computing and leveraging pre-storing intermediate results. On simulated datasets with known ground truth, we show our approach effectively controls type I error and achieves power comparable with commonly used metrics. Furthermore, we demonstrate the performance of our method in evaluating diffusion models on real image datasets with statistical confidence.
Authors: Hao Zhu, Joschka Boedecker
Abstract: We consider the inverse problem of multi-armed bandits (IMAB) that are widely used in neuroscience and psychology research for behavior modelling. We first show that the IMAB problem is not convex in general, but can be relaxed to a convex problem via variable transformation. Based on this result, we propose a two-step sequential heuristic for (approximately) solving the IMAB problem. We discuss a condition where our method provides global solution to the IMAB problem with certificate, as well as approximations to further save computing time. Numerical experiments indicate that our heuristic method is more robust than directly solving the IMAB problem via repeated local optimization, and can achieve the performance of Monte Carlo methods within a significantly decreased running time. We provide the implementation of our method based on CVXPY, which allows straightforward application by users not well versed in convex optimization.
Authors: Anthony Bardou, Patrick Thiran
Abstract: Time-Varying Bayesian Optimization (TVBO) is the go-to framework for optimizing a time-varying, expensive, noisy black-box function. However, most of the solutions proposed so far either rely on unrealistic assumptions on the nature of the objective function or do not offer any theoretical guarantees. We propose the first analysis that asymptotically bounds the cumulative regret of TVBO algorithms under mild and realistic assumptions only. In particular, we provide an algorithm-independent lower regret bound and an upper regret bound that holds for a large class of TVBO algorithms. Based on this analysis, we formulate recommendations for TVBO algorithms and show how an algorithm (BOLT) that follows them performs better than the state-of-the-art of TVBO through experiments on synthetic and real-world problems.
Authors: Gauthier Thurin (CNRS), Kimia Nadjahi (CNRS), Claire Boyer (LMO)
Abstract: Conformal Prediction (CP) is a principled framework for quantifying uncertainty in blackbox learning models, by constructing prediction sets with finite-sample coverage guarantees. Traditional approaches rely on scalar nonconformity scores, which fail to fully exploit the geometric structure of multivariate outputs, such as in multi-output regression or multiclass classification. Recent methods addressing this limitation impose predefined convex shapes for the prediction sets, potentially misaligning with the intrinsic data geometry. We introduce a novel CP procedure handling multivariate score functions through the lens of optimal transport. Specifically, we leverage Monge-Kantorovich vector ranks and quantiles to construct prediction region with flexible, potentially non-convex shapes, better suited to the complex uncertainty patterns encountered in multivariate learning tasks. We prove that our approach ensures finite-sample, distribution-free coverage properties, similar to typical CP methods. We then adapt our method for multi-output regression and multiclass classification, and also propose simple adjustments to generate adaptive prediction regions with asymptotic conditional coverage guarantees. Finally, we evaluate our method on practical regression and classification problems, illustrating its advantages in terms of (conditional) coverage and efficiency.
Authors: Ahmed K. Kadhim, Lei Jiao, Rishad Shafik, Ole-Christoffer Granmo
Abstract: In recent years, text generation tools utilizing Artificial Intelligence (AI) have occasionally been misused across various domains, such as generating student reports or creative writings. This issue prompts plagiarism detection services to enhance their capabilities in identifying AI-generated content. Adversarial attacks are often used to test the robustness of AI-text generated detectors. This work proposes a novel textual adversarial attack on the detection models such as Fast-DetectGPT. The method employs embedding models for data perturbation, aiming at reconstructing the AI generated texts to reduce the likelihood of detection of the true origin of the texts. Specifically, we employ different embedding techniques, including the Tsetlin Machine (TM), an interpretable approach in machine learning for this purpose. By combining synonyms and embedding similarity vectors, we demonstrates the state-of-the-art reduction in detection scores against Fast-DetectGPT. Particularly, in the XSum dataset, the detection score decreased from 0.4431 to 0.2744 AUROC, and in the SQuAD dataset, it dropped from 0.5068 to 0.3532 AUROC.
Authors: Nishikanta Mohanty, Bikash K. Behera, Christopher Ferrie
Abstract: This paper introduces Quantum-SMOTEV2, an advanced variant of the Quantum-SMOTE method, leveraging quantum computing to address class imbalance in machine learning datasets without K-Means clustering. Quantum-SMOTEV2 synthesizes data samples using swap tests and quantum rotation centered around a single data centroid, concentrating on the angular distribution of minority data points and the concept of angular outliers (AOL). Experimental results show significant enhancements in model performance metrics at moderate SMOTE levels (30-36%), which previously required up to 50% with the original method. Quantum-SMOTEV2 maintains essential features of its predecessor (arXiv:2402.17398), such as rotation angle, minority percentage, and splitting factor, allowing for tailored adaptation to specific dataset needs. The method is scalable, utilizing compact swap tests and low depth quantum circuits to accommodate a large number of features. Evaluation on the public Cell-to-Cell Telecom dataset with Random Forest (RF), K-Nearest Neighbours (KNN) Classifier, and Neural Network (NN) illustrates that integrating Angular Outliers modestly boosts classification metrics like accuracy, F1 Score, AUC-ROC, and AUC-PR across different proportions of synthetic data, highlighting the effectiveness of Quantum-SMOTEV2 in enhancing model performance for edge cases.
Authors: Thomas Mortier, Alireza Javanmardi, Yusuf Sale, Eyke H\"ullermeier, Willem Waegeman
Abstract: Conformal prediction has emerged as a widely used framework for constructing valid prediction sets in classification and regression tasks. In this work, we extend the split conformal prediction framework to hierarchical classification, where prediction sets are commonly restricted to internal nodes of a predefined hierarchy, and propose two computationally efficient inference algorithms. The first algorithm returns internal nodes as prediction sets, while the second relaxes this restriction, using the notion of representation complexity, yielding a more general and combinatorial inference problem, but smaller set sizes. Empirical evaluations on several benchmark datasets demonstrate the effectiveness of the proposed algorithms in achieving nominal coverage.
Authors: Maja Pavlovic
Abstract: To be considered reliable, a model must be calibrated so that its confidence in each decision closely reflects its true outcome. In this blogpost we'll take a look at the most commonly used definition for calibration and then dive into a frequently used evaluation measure for model calibration. We'll then cover some of the drawbacks of this measure and how these surfaced the need for additional notions of calibration, which require their own new evaluation measures. This post is not intended to be an in-depth dissection of all works on calibration, nor does it focus on how to calibrate models. Instead, it is meant to provide a gentle introduction to the different notions and their evaluation measures as well as to re-highlight some issues with a measure that is still widely used to evaluate calibration.
Authors: Ruiyu Wang, Yu Yuan, Shizhao Sun, Jiang Bian
Abstract: Creating Computer-Aided Design (CAD) models requires significant expertise and effort. Text-to-CAD, which converts textual descriptions into CAD parametric sequences, is crucial in streamlining this process. Recent studies have utilized ground-truth parametric sequences, known as sequential signals, as supervision to achieve this goal. However, CAD models are inherently multimodal, comprising parametric sequences and corresponding rendered visual objects. Besides,the rendering process from parametric sequences to visual objects is many-to-one. Therefore, both sequential and visual signals are critical for effective training. In this work, we introduce CADFusion, a framework that uses Large Language Models (LLMs) as the backbone and alternates between two training stages: the sequential learning (SL) stage and the visual feedback (VF) stage. In the SL stage, we train LLMs using ground-truth parametric sequences, enabling the generation of logically coherent parametric sequences. In the VF stage, we reward parametric sequences that render into visually preferred objects and penalize those that do not, allowing LLMs to learn how rendered visual objects are perceived and evaluated. These two stages alternate throughout the training, ensuring balanced learning and preserving benefits of both signals. Experiments demonstrate that CADFusion significantly improves performance, both qualitatively and quantitatively.
Authors: Song-Lin Lv, Yu-Yang Chen, Zhi Zhou, Yu-Feng Li, Lan-Zhe Guo
Abstract: Vision-language models (VLMs), such as CLIP, have demonstrated exceptional generalization capabilities and can quickly adapt to downstream tasks through prompt fine-tuning. Unfortunately, in classification tasks involving non-training classes, known as open-vocabulary setting, fine-tuned VLMs often overfit to train classes, resulting in a misalignment between confidence scores and actual accuracy on unseen classes, which significantly undermines their reliability in real-world deployments. Existing confidence calibration methods typically require training parameters or analyzing features from the training dataset, restricting their ability to generalize unseen classes without corresponding train data. Moreover, VLM-specific calibration methods rely solely on text features from train classes as calibration indicators, which inherently limits their ability to calibrate train classes. To address these challenges, we propose an effective multimodal calibration method Contrast-Aware Calibration (CAC). Building on the original CLIP's zero-shot adaptability and the conclusion from empirical analysis that poor intra-class and inter-class discriminative ability on unseen classes is the root cause, we calculate calibration weights based on the contrastive difference between the original and fine-tuned CLIP. This method not only adapts to calibrating unseen classes but also overcomes the limitations of previous VLM calibration methods that could not calibrate train classes. In experiments involving 11 datasets with 5 fine-tuning methods, CAC consistently achieved the best calibration effect on both train and unseen classes without sacrificing accuracy and inference speed.
Authors: A. P. Kryukov, A. P. Demichev, V. A. Ilyin
Abstract: The purpose of this paper is to review the most popular deep learning methods used to analyze astroparticle data obtained with Imaging Atmospheric Cherenkov Telescopes and provide references to the original papers.
Authors: Chuhan Zhang, Cong Wang, Wei Pan, Cosimo Della Santina
Abstract: Inspired by the dynamic coupling of moto-neurons and physical elasticity in animals, this work explores the possibility of generating locomotion gaits by utilizing physical oscillations in a soft snake by means of a low-level spiking neural mechanism. To achieve this goal, we introduce the Double Threshold Spiking neuron model with adjustable thresholds to generate varied output patterns. This neuron model can excite the natural dynamics of soft robotic snakes, and it enables distinct movements, such as turning or moving forward, by simply altering the neural thresholds. Finally, we demonstrate that our approach, termed SpikingSoft, naturally pairs and integrates with reinforcement learning. The high-level agent only needs to adjust the two thresholds to generate complex movement patterns, thus strongly simplifying the learning of reactive locomotion. Simulation results demonstrate that the proposed architecture significantly enhances the performance of the soft snake robot, enabling it to achieve target objectives with a 21.6% increase in success rate, a 29% reduction in time to reach the target, and smoother movements compared to the vanilla reinforcement learning controllers or Central Pattern Generator controller acting in torque space.
Authors: F. Stricker, J. A. Peregrina, D. Bermbach, C. Zirpins
Abstract: Federated Learning (FL) is an upcoming technology that is increasingly applied in real-world applications. Early applications focused on cross-device scenarios, where many participants with limited resources train machine learning (ML) models together, e.g., in the case of Google's GBoard. Contrarily, cross-silo scenarios have only few participants but with many resources, e.g., in the healthcare domain. Despite such early efforts, FL is still rarely used in practice and best practices are, hence, missing. For new applications, in our case inter-organizational cross-silo applications, overcoming this lack of role models is a significant challenge. In order to ease the use of FL in real-world cross-silo applications, we here propose a scenario-based architecture for the practical use of FL in the context of multiple companies collaborating to improve the quality of their ML models. The architecture emphasizes the collaboration between the participants and the FL server and extends basic interactions with domain-specific features. First, it combines governance with authentication, creating an environment where only trusted participants can join. Second, it offers traceability of governance decisions and tracking of training processes, which are also crucial in a production environment. Beyond presenting the architectural design, we analyze requirements for the real-world use of FL and evaluate the architecture with a scenario-based analysis method.
Authors: Ioannis Reklos, Jacopo de Berardinis, Elena Simperl, Albert Mero\~no-Pe\~nuela
Abstract: Knowledge Graphs (KGs) store human knowledge in the form of entities (nodes) and relations, and are used extensively in various applications. KG embeddings are an effective approach to addressing tasks like knowledge discovery, link prediction, and reasoning. This is often done by allocating and learning embedding tables for all or a subset of the entities. As this scales linearly with the number of entities, learning embedding models in real-world KGs with millions of nodes can be computationally intractable. To address this scalability problem, our model, PathE, only allocates embedding tables for relations (which are typically orders of magnitude fewer than the entities) and requires less than 25% of the parameters of previous parameter efficient methods. Rather than storing entity embeddings, we learn to compute them by leveraging multiple entity-relation paths to contextualise individual entities within triples. Evaluated on four benchmarks, PathE achieves state-of-the-art performance in relation prediction, and remains competitive in link prediction on path-rich KGs while training on consumer-grade hardware. We perform ablation experiments to test our design choices and analyse the sensitivity of the model to key hyper-parameters. PathE is efficient and cost-effective for relationally diverse and well-connected KGs commonly found in real-world applications.
Authors: Saul Santos, Ant\'onio Farinhas, Daniel C. McNamee, Andr\'e F. T. Martins
Abstract: Current video-language models struggle with long-video understanding due to limited context lengths and reliance on sparse frame subsampling, often leading to information loss. This paper introduces $\infty$-Video, which can process arbitrarily long videos through a continuous-time long-term memory (LTM) consolidation mechanism. Our framework augments video Q-formers by allowing them to process unbounded video contexts efficiently and without requiring additional training. Through continuous attention, our approach dynamically allocates higher granularity to the most relevant video segments, forming "sticky" memories that evolve over time. Experiments with Video-LLaMA and VideoChat2 demonstrate improved performance in video question-answering tasks, showcasing the potential of continuous-time LTM mechanisms to enable scalable and training-free comprehension of long videos.
Authors: Paul Fuchs, Micha{\l} Sanocki, Julija Zavadlav
Abstract: Graph Neural Network (GNN) potentials relying on chemical locality offer near-quantum mechanical accuracy at significantly reduced computational costs. By propagating local information to distance particles, Message-passing neural networks (MPNNs) extend the locality concept to model interactions beyond their local neighborhood. Still, this locality precludes modeling long-range effects, such as charge transfer, electrostatic interactions, and dispersion effects, which are critical to adequately describe many real-world systems. In this work, we propose the Charge Equilibration Layer for Long-range Interactions (CELLI) to address the challenging modeling of non-local interactions and the high computational cost of MPNNs. This novel architecture generalizes the fourth-generation high-dimensional neural network (4GHDNN) concept, integrating the charge equilibration (Qeq) method into a model-agnostic building block for modern equivariant GNN potentials. A series of benchmarks show that CELLI can extend the strictly local Allegro architecture to model highly non-local interactions and charge transfer. Our architecture generalizes to diverse datasets and large structures, achieving an accuracy comparable to MPNNs at about twice the computational efficiency.
Authors: Xuan Shen, Yizhou Wang, Xiangxi Shi, Yanzhi Wang, Pu Zhao, Jiuxiang Gu
Abstract: Chain-of-Thought (CoT) reasoning has become a powerful framework for improving complex problem-solving capabilities in Multimodal Large Language Models (MLLMs). However, the verbose nature of textual reasoning introduces significant inefficiencies. In this work, we propose $\textbf{Heima}$ (as hidden llama), an efficient reasoning framework that leverages reasoning CoTs at hidden latent space. We design the Heima Encoder to condense each intermediate CoT into a compact, higher-level hidden representation using a single thinking token, effectively minimizing verbosity and reducing the overall number of tokens required during the reasoning process. Meanwhile, we design corresponding Heima Decoder with traditional Large Language Models (LLMs) to adaptively interpret the hidden representations into variable-length textual sequence, reconstructing reasoning processes that closely resemble the original CoTs. Experimental results across diverse reasoning MLLM benchmarks demonstrate that Heima model achieves higher generation efficiency while maintaining or even better zero-shot task accuracy. Moreover, the effective reconstruction of multimodal reasoning processes with Heima Decoder validates both the robustness and interpretability of our approach.
Authors: Leonardo Di Nino, Sergio Barbarossa, Paolo Di Lorenzo
Abstract: The aim of this paper is to propose a novel framework to infer the sheaf Laplacian, including the topology of a graph and the restriction maps, from a set of data observed over the nodes of a graph. The proposed method is based on sheaf theory, which represents an important generalization of graph signal processing. The learning problem aims to find the sheaf Laplacian that minimizes the total variation of the observed data, where the variation over each edge is also locally minimized by optimizing the associated restriction maps. Compared to alternative methods based on semidefinite programming, our solution is significantly more numerically efficient, as all its fundamental steps are resolved in closed form. The method is numerically tested on data consisting of vectors defined over subspaces of varying dimensions at each node. We demonstrate how the resulting graph is influenced by two key factors: the cross-correlation and the dimensionality difference of the data residing on the graph's nodes.
Authors: Hansheng Jiang, Chunlin Sun, Zuo-Jun Max Shen, Shunan Jiang
Abstract: We consider a network inventory problem motivated by one-way, on-demand vehicle sharing services. Due to uncertainties in both demand and returns, as well as a fixed number of rental units across an $n$-location network, the service provider must periodically reposition vehicles to match supply with demand spatially while minimizing costs. The optimal repositioning policy under a general $n$-location network is intractable without knowing the optimal value function. We introduce the best base-stock repositioning policy as a generalization of the classical inventory control policy to $n$ dimensions, and establish its asymptotic optimality in two distinct limiting regimes under general network structures. We present reformulations to efficiently compute this best base-stock policy in an offline setting with pre-collected data. In the online setting, we show that a natural Lipschitz-bandit approach achieves a regret guarantee of $\widetilde{O}(T^{\frac{n}{n+1}})$, which suffers from the exponential dependence on $n$. We illustrate the challenges of learning with censored data in networked systems through a regret lower bound analysis and by demonstrating the suboptimality of alternative algorithmic approaches. Motivated by these challenges, we propose an Online Gradient Repositioning algorithm that relies solely on censored demand. Under a mild cost-structure assumption, we prove that it attains an optimal regret of $O(n^{2.5} \sqrt{T})$, which matches the regret lower bound in $T$ and achieves only polynomial dependence on $n$. The key algorithmic innovation involves proposing surrogate costs to disentangle intertemporal dependencies and leveraging dual solutions to find the gradient of policy change. Numerical experiments demonstrate the effectiveness of our proposed methods.
Authors: Wei Liu, Yangyang Xu
Abstract: Many real-world problems, such as those with fairness constraints, involve complex expectation constraints and large datasets, necessitating the design of efficient stochastic methods to solve them. Most existing research focuses on cases with no {constraint} or easy-to-project constraints or deterministic constraints. In this paper, we consider nonconvex nonsmooth stochastic optimization problems with expectation constraints, for which we build a novel exact penalty model. We first show the relationship between the penalty model and the original problem. Then on solving the penalty problem, we present a single-loop SPIDER-type stochastic subgradient method, which utilizes the subgradients of both the objective and constraint functions, as well as the constraint function value at each iteration. Under certain regularity conditions (weaker than Slater-type constraint qualification or strong feasibility assumed in existing works), we establish an iteration complexity result of $O(\epsilon^{-4})$ to reach a near-$\epsilon$ stationary point of the penalized problem in expectation, matching the lower bound for such tasks. Building on the exact penalization, an $(\epsilon,\epsilon)$-KKT point of the original problem is obtained. For a few scenarios, our complexity of either the {objective} sample subgradient or the constraint sample function values can be lower than the state-of-the-art results by a factor of $\epsilon^{-2}$. Moreover, on solving two fairness-constrained problems, our method is significantly (up to 466 times) faster than the state-of-the-art algorithms, including switching subgradient method and inexact proximal point methods.
Authors: Abdulaziz Alghamdi, David Mohaisen
Abstract: \begin{abstract} This paper comprehensively analyzes privacy policies in AR/VR applications, leveraging BERT, a state-of-the-art text classification model, to evaluate the clarity and thoroughness of these policies. By comparing the privacy policies of AR/VR applications with those of free and premium websites, this study provides a broad perspective on the current state of privacy practices within the AR/VR industry. Our findings indicate that AR/VR applications generally offer a higher percentage of positive segments than free content but lower than premium websites. The analysis of highlighted segments and words revealed that AR/VR applications strategically emphasize critical privacy practices and key terms. This enhances privacy policies' clarity and effectiveness.
Authors: BaoLinh Tran, Van Vu
Abstract: The matrix recovery (completion) problem, a central problem in data science and theoretical computer science, is to recover a matrix $A$ from a relatively small sample of entries. While such a task is impossible in general, it has been shown that one can recover $A$ exactly in polynomial time, with high probability, from a random subset of entries, under three (basic and necessary) assumptions: (1) the rank of $A$ is very small compared to its dimensions (low rank), (2) $A$ has delocalized singular vectors (incoherence), and (3) the sample size is sufficiently large. There are many different algorithms for the task, including convex optimization by Candes, Tao and Recht (2009), alternating projection by Hardt and Wooters (2014) and low rank approximation with gradient descent by Keshavan, Montanari and Oh (2009, 2010). In applications, it is more realistic to assume that data is noisy. In this case, these approaches provide an approximate recovery with small root mean square error. However, it is hard to transform such approximate recovery to an exact one. Recently, results by Abbe et al. (2017) and Bhardwaj et al. (2023) concerning approximation in the infinity norm showed that we can achieve exact recovery even in the noisy case, given that the ground matrix has bounded precision. Beyond the three basic assumptions above, they required either the condition number of $A$ is small (Abbe et al.) or the gap between consecutive singular values is large (Bhardwaj et al.). In this paper, we remove these extra spectral assumptions. As a result, we obtain a simple algorithm for exact recovery in the noisy case, under only three basic assumptions. This is the first such algorithm. To analyse the algorithm, we introduce a contour integration argument which is totally different from all previous methods and may be of independent interest.
Authors: Arsenii Gavrikov, Juli\'an Garc\'ia Pardi\~nas, Alberto Garfagnini
Abstract: Ensuring reliable data collection in large-scale particle physics experiments demands Data Quality Monitoring (DQM) procedures to detect possible detector malfunctions and preserve data integrity. Traditionally, this resource-intensive task has been handled by human shifters that struggle with frequent changes in operational conditions. We present novel, interpretable, robust, and scalable DQM algorithms designed to automate anomaly detection in time-dependent settings. Our approach constructs evolving histogram templates with built-in uncertainties, featuring both a statistical variant - extending the classical Exponentially Weighted Moving Average (EWMA) - and a machine learning (ML)-enhanced version that leverages a transformer encoder for improved adaptability. Experimental validations on synthetic datasets demonstrate the high accuracy, adaptability, and interpretability of these methods, with the statistical variant being commissioned in the LHCb experiment at the Large Hadron Collider, underscoring its real-world impact. The code used in this study is available at https://github.com/ArseniiGav/DINAMO.
Authors: Amogh Joshi, Sourav Sanyal, Kaushik Roy
Abstract: The integration of human-intuitive interactions into autonomous systems has been limited. Traditional Natural Language Processing (NLP) systems struggle with context and intent understanding, severely restricting human-robot interaction. Recent advancements in Large Language Models (LLMs) have transformed this dynamic, allowing for intuitive and high-level communication through speech and text, and bridging the gap between human commands and robotic actions. Additionally, autonomous navigation has emerged as a central focus in robotics research, with artificial intelligence (AI) increasingly being leveraged to enhance these systems. However, existing AI-based navigation algorithms face significant challenges in latency-critical tasks where rapid decision-making is critical. Traditional frame-based vision systems, while effective for high-level decision-making, suffer from high energy consumption and latency, limiting their applicability in real-time scenarios. Neuromorphic vision systems, combining event-based cameras and spiking neural networks (SNNs), offer a promising alternative by enabling energy-efficient, low-latency navigation. Despite their potential, real-world implementations of these systems, particularly on physical platforms such as drones, remain scarce. In this work, we present Neuro-LIFT, a real-time neuromorphic navigation framework implemented on a Parrot Bebop2 quadrotor. Leveraging an LLM for natural language processing, Neuro-LIFT translates human speech into high-level planning commands which are then autonomously executed using event-based neuromorphic vision and physics-driven planning. Our framework demonstrates its capabilities in navigating in a dynamic environment, avoiding obstacles, and adapting to human instructions in real-time.
Authors: Orion Weller, Benjamin Chang, Eugene Yang, Mahsa Yarmohammadi, Sam Barham, Sean MacAvaney, Arman Cohan, Luca Soldaini, Benjamin Van Durme, Dawn Lawrie
Abstract: Retrieval systems generally focus on web-style queries that are short and underspecified. However, advances in language models have facilitated the nascent rise of retrieval models that can understand more complex queries with diverse intents. However, these efforts have focused exclusively on English; therefore, we do not yet understand how they work across languages. We introduce mFollowIR, a multilingual benchmark for measuring instruction-following ability in retrieval models. mFollowIR builds upon the TREC NeuCLIR narratives (or instructions) that span three diverse languages (Russian, Chinese, Persian) giving both query and instruction to the retrieval models. We make small changes to the narratives and isolate how well retrieval models can follow these nuanced changes. We present results for both multilingual (XX-XX) and cross-lingual (En-XX) performance. We see strong cross-lingual performance with English-based retrievers that trained using instructions, but find a notable drop in performance in the multilingual setting, indicating that more work is needed in developing data for instruction-based multilingual retrievers.
Authors: David Li, Anvar Kurmukov, Mikhail Goncharov, Roman Sokolov, Mikhail Belyaev
Abstract: Recent advances in deep learning have shown that learning robust feature representations is critical for the success of many computer vision tasks, including medical image segmentation. In particular, both transformer and convolutional-based architectures have benefit from leveraging pretext tasks for pretraining. However, the adoption of pretext tasks in 3D medical imaging has been less explored and remains a challenge, especially in the context of learning generalizable feature representations. We propose a novel pretraining strategy using diffusion models with anatomical guidance, tailored to the intricacies of 3D medical image data. We introduce an auxiliary diffusion process to pretrain a model that produce generalizable feature representations, useful for a variety of downstream segmentation tasks. We employ an additional model that predicts 3D universal body-part coordinates, providing guidance during the diffusion process and improving spatial awareness in generated representations. This approach not only aids in resolving localization inaccuracies but also enriches the model's ability to understand complex anatomical structures. Empirical validation on a 13-class organ segmentation task demonstrate the effectiveness of our pretraining technique. It surpasses existing restorative pretraining methods in 3D medical image segmentation by $7.5\%$, and is competitive with the state-of-the-art contrastive pretraining approach, achieving an average Dice coefficient of 67.8 in a non-linear evaluation scenario.
Authors: Roberto-Rafael Maura-Rivero, Marc Lanctot, Francesco Visin, Kate Larson
Abstract: Reinforcement Learning from Human Feedback (RLHF), the standard for aligning Large Language Models (LLMs) with human values, is known to fail to satisfy properties that are intuitively desirable, such as respecting the preferences of the majority \cite{ge2024axioms}. To overcome these issues, we propose the use of a probabilistic Social Choice rule called \emph{maximal lotteries} as a replacement for RLHF. We show that a family of alignment techniques, namely Nash Learning from Human Feedback (NLHF) \cite{munos2023nash} and variants, approximate maximal lottery outcomes and thus inherit its beneficial properties. We confirm experimentally that our proposed methodology handles situations that arise when working with preferences more robustly than standard RLHF, including supporting the preferences of the majority, providing principled ways of handling non-transitivities in the preference data, and robustness to irrelevant alternatives. This results in systems that better incorporate human values and respect human intentions.
Authors: Halil Ibrahim Aysel, Xiaohao Cai, Adam Prugel-Bennett
Abstract: Concept-based explanation methods, such as concept bottleneck models (CBMs), aim to improve the interpretability of machine learning models by linking their decisions to human-understandable concepts, under the critical assumption that such concepts can be accurately attributed to the network's feature space. However, this foundational assumption has not been rigorously validated, mainly because the field lacks standardised metrics and benchmarks to assess the existence and spatial alignment of such concepts. To address this, we propose three metrics: the concept global importance metric, the concept existence metric, and the concept location metric, including a technique for visualising concept activations, i.e., concept activation mapping. We benchmark post-hoc CBMs to illustrate their capabilities and challenges. Through qualitative and quantitative experiments, we demonstrate that, in many cases, even the most important concepts determined by post-hoc CBMs are not present in input images; moreover, when they are present, their saliency maps fail to align with the expected regions by either activating across an entire object or misidentifying relevant concept-specific regions. We analyse the root causes of these limitations, such as the natural correlation of concepts. Our findings underscore the need for more careful application of concept-based explanation techniques especially in settings where spatial interpretability is critical.
Authors: Jierui Zuo, Hanzhang Qin
Abstract: We provide a new online learning algorithm for tackling the Multinomial Logit Bandit (MNL-Bandit) problem. Despite the challenges posed by the combinatorial nature of the MNL model, we develop a novel Upper Confidence Bound (UCB)-based method that achieves Pareto optimality by balancing regret minimization and estimation error of the assortment revenues and the MNL parameters. We develop theoretical guarantees characterizing the tradeoff between regret and estimation error for the MNL-Bandit problem through information-theoretic bounds, and propose a modified UCB algorithm that incorporates forced exploration to improve parameter estimation accuracy while maintaining low regret. Our analysis sheds critical insights into how to optimally balance the collected revenues and the treatment estimation in dynamic assortment optimization.
Authors: Sebastiano Ariosto
Abstract: Deep Neural Networks (DNNs) excel at many tasks, often rivaling or surpassing human performance. Yet their internal processes remain elusive, frequently described as "black boxes." While performance can be refined experimentally, achieving a fundamental grasp of their inner workings is still a challenge. Statistical Mechanics has long tackled computational problems, and this thesis applies physics-based insights to understand DNNs via three complementary approaches. First, by averaging over data, we derive an asymptotic bound on generalization that depends solely on the size of the last layer, rather than on the total number of parameters -- revealing how deep architectures process information differently across layers. Second, adopting a data-dependent viewpoint, we explore a finite-width thermodynamic limit beyond the infinite-width regime. This leads to: (i) a closed-form expression for the generalization error in a finite-width one-hidden-layer network (regression task); (ii) an approximate partition function for deeper architectures; and (iii) a link between deep networks in this thermodynamic limit and Student's t-processes. Finally, from a task-explicit perspective, we present a preliminary analysis of how DNNs interact with a controlled dataset, investigating whether they truly internalize its structure -- collapsing to the teacher -- or merely memorize it. By understanding when a network must learn data structure rather than just memorize, it sheds light on fostering meaningful internal representations. In essence, this thesis leverages the synergy between Statistical Physics and Machine Learning to illuminate the inner behavior of DNNs.
Authors: Amritendu Mukherjee, Dipanwita Sinha Mukherjee, Parthasarathy Ramachandran
Abstract: Training a neural network for pixel based classification task using low resolution Landsat images is difficult as the size of the training data is usually small due to less number of available pixels that represent a single class without any mixing with other classes. Due to this scarcity of training data, neural network may not be able to attain expected level of accuracy. This limitation could be overcome using a generative network that aims to generate synthetic data having the same distribution as the sample data with which it is trained. In this work, we have proposed a methodology for improving the performance of ANN classifier to identify built-up pixels in the Landsat$7$ image with the help of developing a simple GAN architecture that could generate synthetic training pixels when trained using original set of sample built-up pixels. To ensure that the marginal and joint distributions of all the bands corresponding to the generated and original set of pixels are indistinguishable, non-parametric Kolmogorov Smirnov Test and Ball Divergence based Equality of Distributions Test have been performed respectively. It has been observed that the overall accuracy and kappa coefficient of the ANN model for built-up classification have continuously improved from $0.9331$ to $0.9983$ and $0.8277$ to $0.9958$ respectively, with the inclusion of generated sets of built-up pixels to the original one.
Authors: Zhiyao Xu, Dan Zhao, Qingsong Zou, Jingyu Xiao, Yong Jiang, Zhenhui Yuan, Qing Li
Abstract: In recent years, as smart home systems have become more widespread, security concerns within these environments have become a growing threat. Currently, most smart home security solutions, such as anomaly detection and behavior prediction models, are trained using fixed datasets that are precollected. However, the process of dataset collection is time-consuming and lacks the flexibility needed to adapt to the constantly evolving smart home environment. Additionally, the collection of personal data raises significant privacy concerns for users. Lately, large language models (LLMs) have emerged as a powerful tool for a wide range of tasks across diverse application domains, thanks to their strong capabilities in natural language processing, reasoning, and problem-solving. In this paper, we propose an LLM-based synthetic dataset generation IoTGen framework to enhance the generalization of downstream smart home intelligent models. By generating new synthetic datasets that reflect changes in the environment, smart home intelligent models can be retrained to overcome the limitations of fixed and outdated data, allowing them to better align with the dynamic nature of real-world home environments. Specifically, we first propose a Structure Pattern Perception Compression (SPPC) method tailored for IoT behavior data, which preserves the most informative content in the data while significantly reducing token consumption. Then, we propose a systematic approach to create prompts and implement data generation to automatically generate IoT synthetic data with normative and reasonable properties, assisting task models in adaptive training to improve generalization and real-world performance.
Authors: Edward Storey, Naomi Harte, Peter Bell
Abstract: Self-supervised learning (SSL) is used in deep learning to train on large datasets without the need for expensive labelling of the data. Recently, large Automatic Speech Recognition (ASR) models such as XLS-R have utilised SSL to train on over one hundred different languages simultaneously. However, deeper investigation shows that the bulk of the training data for XLS-R comes from a small number of languages. Biases learned through SSL have been shown to exist in multiple domains, but language bias in multilingual SSL ASR has not been thoroughly examined. In this paper, we utilise the Lottery Ticket Hypothesis (LTH) to identify language-specific subnetworks within XLS-R and test the performance of these subnetworks on a variety of different languages. We are able to show that when fine-tuning, XLS-R bypasses traditional linguistic knowledge and builds only on weights learned from the languages with the largest data contribution to the pretraining data.
Authors: Unai Fischer-Abaigar, Christoph Kern, Juan Carlos Perdomo
Abstract: Machine learning is increasingly used in government programs to identify and support the most vulnerable individuals, prioritizing assistance for those at greatest risk over optimizing aggregate outcomes. This paper examines the welfare impacts of prediction in equity-driven contexts, and how they compare to other policy levers, such as expanding bureaucratic capacity. Through mathematical models and a real-world case study on long-term unemployment amongst German residents, we develop a comprehensive understanding of the relative effectiveness of prediction in surfacing the worst-off. Our findings provide clear analytical frameworks and practical, data-driven tools that empower policymakers to make principled decisions when designing these systems.
Authors: Frederik Hytting J{\o}rgensen, Luigi Gresele, Sebastian Weichwald
Abstract: Causal Bayesian networks are 'causal' models since they make predictions about interventional distributions. To connect such causal model predictions to real-world outcomes, we must determine which actions in the world correspond to which interventions in the model. For example, to interpret an action as an intervention on a treatment variable, the action will presumably have to a) change the distribution of treatment in a way that corresponds to the intervention, and b) not change other aspects, such as how the outcome depends on the treatment; while the marginal distributions of some variables may change as an effect. We introduce a formal framework to make such requirements for different interpretations of actions as interventions precise. We prove that the seemingly natural interpretation of actions as interventions is circular: Under this interpretation, every causal Bayesian network that correctly models the observational distribution is trivially also interventionally valid, and no action yields empirical data that could possibly falsify such a model. We prove an impossibility result: No interpretation exists that is non-circular and simultaneously satisfies a set of natural desiderata. Instead, we examine non-circular interpretations that may violate some desiderata and show how this may in turn enable the falsification of causal models. By rigorously examining how a causal Bayesian network could be a 'causal' model of the world instead of merely a mathematical object, our formal framework contributes to the conceptual foundations of causal representation learning, causal discovery, and causal abstraction, while also highlighting some limitations of existing approaches.
Authors: Emily Wenger, Yoed Kenett
Abstract: Numerous powerful large language models (LLMs) are now available for use as writing support tools, idea generators, and beyond. Although these LLMs are marketed as helpful creative assistants, several works have shown that using an LLM as a creative partner results in a narrower set of creative outputs. However, these studies only consider the effects of interacting with a single LLM, begging the question of whether such narrowed creativity stems from using a particular LLM -- which arguably has a limited range of outputs -- or from using LLMs in general as creative assistants. To study this question, we elicit creative responses from humans and a broad set of LLMs using standardized creativity tests and compare the population-level diversity of responses. We find that LLM responses are much more similar to other LLM responses than human responses are to each other, even after controlling for response structure and other key variables. This finding of significant homogeneity in creative outputs across the LLMs we evaluate adds a new dimension to the ongoing conversation about creativity and LLMs. If today's LLMs behave similarly, using them as a creative partners -- regardless of the model used -- may drive all users towards a limited set of "creative" outputs.
Authors: S\"oren Christensen, Claudia Strauch, Lukas Trottner
Abstract: We introduce a new class of generative diffusion models that, unlike conventional denoising diffusion models, achieve a time-homogeneous structure for both the noising and denoising processes, allowing the number of steps to adaptively adjust based on the noise level. This is accomplished by conditioning the forward process using Doob's $h$-transform, which terminates the process at a suitable sampling distribution at a random time. The model is particularly well suited for generating data with lower intrinsic dimensions, as the termination criterion simplifies to a first-hitting rule. A key feature of the model is its adaptability to the target data, enabling a variety of downstream tasks using a pre-trained unconditional generative model. These tasks include natural conditioning through appropriate initialization of the denoising process and classification of noisy data.
Authors: Dominik Wagner, Alexander Churchill, Siddarth Sigtia, Erik Marchi
Abstract: In this work, we present and evaluate SELMA, a Speech-Enabled Language Model for virtual Assistant interactions that integrates audio and text as inputs to a Large Language Model (LLM). SELMA is designed to handle three primary and two auxiliary tasks related to interactions with virtual assistants simultaneously within a single end-to-end model. We employ low-rank adaptation modules for parameter-efficient training of both the audio encoder and the LLM. Additionally, we implement a feature pooling strategy enabling the system to recognize global patterns and improve accuracy on tasks less reliant on individual sequence elements. Experimental results on Voice Trigger (VT) detection, Device-Directed Speech Detection (DDSD), and Automatic Speech Recognition (ASR), demonstrate that our approach both simplifies the typical input processing pipeline of virtual assistants significantly and also improves performance compared to dedicated models for each individual task. SELMA yields relative Equal-Error Rate improvements of 64% on the VT detection task, and 22% on DDSD, while also achieving word error rates close to the baseline.
Authors: Weimin Zhou
Abstract: It is widely accepted that the Bayesian ideal observer (IO) should be used to guide the objective assessment and optimization of medical imaging systems. The IO employs complete task-specific information to compute test statistics for making inference decisions and performs optimally in signal detection tasks. However, the IO test statistic typically depends non-linearly on the image data and cannot be analytically determined. The ideal linear observer, known as the Hotelling observer (HO), can sometimes be used as a surrogate for the IO. However, when image data are high dimensional, HO computation can be difficult. Efficient channels that can extract task-relevant features have been investigated to reduce the dimensionality of image data to approximate IO and HO performance. This work proposes a novel method for generating efficient channels by use of the gradient of a Lagrangian-based loss function that was designed to learn the HO. The generated channels are referred to as the Lagrangian-gradient (L-grad) channels. Numerical studies are conducted that consider binary signal detection tasks involving various backgrounds and signals. It is demonstrated that channelized HO (CHO) using L-grad channels can produce significantly better signal detection performance compared to the CHO using PLS channels. Moreover, it is shown that the proposed L-grad method can achieve significantly lower computation time compared to the PLS method.
Authors: Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Cand\`es, Tatsunori Hashimoto
Abstract: Test-time scaling is a promising new approach to language modeling that uses extra test-time compute to improve performance. Recently, OpenAI's o1 model showed this capability but did not publicly share its methodology, leading to many replication efforts. We seek the simplest approach to achieve test-time scaling and strong reasoning performance. First, we curate a small dataset s1K of 1,000 questions paired with reasoning traces relying on three criteria we validate through ablations: difficulty, diversity, and quality. Second, we develop budget forcing to control test-time compute by forcefully terminating the model's thinking process or lengthening it by appending "Wait" multiple times to the model's generation when it tries to end. This can lead the model to double-check its answer, often fixing incorrect reasoning steps. After supervised finetuning the Qwen2.5-32B-Instruct language model on s1K and equipping it with budget forcing, our model s1 exceeds o1-preview on competition math questions by up to 27% (MATH and AIME24). Further, scaling s1 with budget forcing allows extrapolating beyond its performance without test-time intervention: from 50% to 57% on AIME24. Our model, data, and code are open-source at https://github.com/simplescaling/s1.
Authors: Mustafa O. Karabag, Ufuk Topcu
Abstract: Large language model-based (LLM-based) agents have become common in settings that include non-cooperative parties. In such settings, agents' decision-making needs to conceal information from their adversaries, reveal information to their cooperators, and infer information to identify the other agents' characteristics. To investigate whether LLMs have these information control and decision-making capabilities, we make LLM agents play the language-based hidden-identity game, The Chameleon. In the game, a group of non-chameleon agents who do not know each other aim to identify the chameleon agent without revealing a secret. The game requires the aforementioned information control capabilities both as a chameleon and a non-chameleon. The empirical results show that while non-chameleon LLM agents identify the chameleon, they fail to conceal the secret from the chameleon, and their winning probability is far from the levels of even trivial strategies. To formally explain this behavior, we give a theoretical analysis for a spectrum of strategies, from concealing to revealing, and provide bounds on the non-chameleons' winning probability. Based on the empirical results and theoretical analysis of different strategies, we deduce that LLM-based non-chameleon agents reveal excessive information to agents of unknown identities. Our results point to a weakness of contemporary LLMs, including GPT-4, GPT-4o, Gemini 1.5, and Claude 3.5 Sonnet, in strategic interactions.
Authors: Ken M. Nakanishi
Abstract: The maximum element of the vector output by the Softmax function approaches zero as the input vector size increases. Transformer-based language models rely on Softmax to compute attention scores, causing the attention distribution to flatten as the context size grows. This reduces the model's ability to prioritize key information effectively and potentially limits its length generalization. To address this problem, we propose Scalable-Softmax (SSMax), which replaces Softmax in scenarios where the input vector size varies. SSMax can be seamlessly integrated into existing Transformer-based architectures. Experimental results in language modeling show that models using SSMax not only achieve faster loss reduction during pretraining but also significantly improve performance in long contexts and key information retrieval. Furthermore, an analysis of attention scores reveals that SSMax enables the model to focus attention on key information even in long contexts. Additionally, although models that use SSMax from the beginning of pretraining achieve better length generalization, those that have already started pretraining can still gain some of this ability by replacing Softmax in the attention layers with SSMax, either during or after pretraining.
Authors: Pat Pataranutaporn, Nattavudh Powdthavee, Pattie Maes
Abstract: Surnames often convey implicit markers of social status, wealth, and lineage, shaping perceptions in ways that can perpetuate systemic biases. This study investigates whether and how surnames influence AI-driven decision-making, focusing on their effects across key areas such as hiring recommendations, leadership appointments, and loan approvals. Drawing on 600 surnames from the United States and Thailand, countries with differing sociohistorical dynamics and surname conventions, we categorize names into Rich, Legacy, Normal, and phonetically similar Variant groups. Our findings reveal that elite surnames consistently predict AI-generated perceptions of power, intelligence, and wealth, leading to significant consequences for decisions in high-stakes situations. Mediation analysis highlights perceived intelligence as a crucial pathway through which surname biases operate. Providing objective qualifications alongside the surnames reduces, but does not eliminate, these biases, especially in contexts with uniformly low credentials. These results call for fairness-aware algorithms and robust policy interventions to mitigate the reinforcement of inherited inequalities by AI systems. Our work also urges a reexamination of algorithmic accountability and its societal impact, particularly in systems designed for meritocratic outcomes.
Authors: Tasuku Soma, Khashayar Gatmiry, Sharut Gupta, Stefanie Jegelka
Abstract: Distributionally robust optimization (DRO) can improve the robustness and fairness of learning methods. In this paper, we devise stochastic algorithms for a class of DRO problems including group DRO, subpopulation fairness, and empirical conditional value at risk (CVaR) optimization. Our new algorithms achieve faster convergence rates than existing algorithms for multiple DRO settings. We also provide a new information-theoretic lower bound that implies our bounds are tight for group DRO. Empirically, too, our algorithms outperform known methods.
Authors: Antoine de Mathelin, Fran\c{c}ois Deheeger, Mathilde Mougeot, Nicolas Vayatis
Abstract: This paper deals with uncertainty quantification and out-of-distribution detection in deep learning using Bayesian and ensemble methods. It proposes a practical solution to the lack of prediction diversity observed recently for standard approaches when used out-of-distribution (Ovadia et al., 2019; Liu et al., 2021). Considering that this issue is mainly related to a lack of weight diversity, we claim that standard methods sample in "over-restricted" regions of the weight space due to the use of "over-regularization" processes, such as weight decay and zero-mean centered Gaussian priors. We propose to solve the problem by adopting the maximum entropy principle for the weight distribution, with the underlying idea to maximize the weight diversity. Under this paradigm, the epistemic uncertainty is described by the weight distribution of maximal entropy that produces neural networks "consistent" with the training observations. Considering stochastic neural networks, a practical optimization is derived to build such a distribution, defined as a trade-off between the average empirical risk and the weight distribution entropy. We develop a novel weight parameterization for the stochastic model, based on the singular value decomposition of the neural network's hidden representations, which enables a large increase of the weight entropy for a small empirical risk penalization. We provide both theoretical and numerical results to assess the efficiency of the approach. In particular, the proposed algorithm appears in the top three best methods in all configurations of an extensive out-of-distribution detection benchmark including more than thirty competitors.
Authors: Ahmad Beirami, Alekh Agarwal, Jonathan Berant, Alexander D'Amour, Jacob Eisenstein, Chirag Nagpal, Ananda Theertha Suresh
Abstract: A simple and effective method for the inference-time alignment of generative models is the best-of-$n$ policy, where $n$ samples are drawn from a reference policy, ranked based on a reward function, and the highest ranking one is selected. A commonly used analytical expression in the literature claims that the KL divergence between the best-of-$n$ policy and the reference policy is equal to $\log (n) - (n-1)/n.$ We disprove the validity of this claim, and show that it is an upper bound on the actual KL divergence. We also explore the tightness of this upper bound in different regimes, and propose a new estimator for the KL divergence and empirically show that it provides a tight approximation. We also show that the win rate of the best-of-$n$ policy against the reference policy is upper bounded by $n/(n+1)$ and derive bounds on the tightness of this characterization. We conclude with analyzing the tradeoffs between win rate and KL divergence of the best-of-$n$ alignment policy, which demonstrate that very good tradeoffs are achievable with $n < 1000$.
Authors: Xinyu Tang, Ashwinee Panda, Milad Nasr, Saeed Mahloujifar, Prateek Mittal
Abstract: Differentially private stochastic gradient descent (DP-SGD) allows models to be trained in a privacy-preserving manner, but has proven difficult to scale to the era of foundation models. We introduce DP-ZO, a private fine-tuning framework for large language models by privatizing zeroth order optimization methods. A key insight into the design of our method is that the direction of the gradient in the zeroth-order optimization we use is random and the only information from training data is the step size, i.e., a scalar. Therefore, we only need to privatize the scalar step size, which is memory-efficient. DP-ZO provides a strong privacy-utility trade-off across different tasks, and model sizes that are comparable to DP-SGD in $(\varepsilon,\delta)$-DP. Notably, DP-ZO possesses significant advantages over DP-SGD in memory efficiency, and obtains higher utility in $\varepsilon$-DP when using the Laplace mechanism.
Authors: Sebastian Bordt, Eric Raidl, Ulrike von Luxburg
Abstract: In the rapidly growing literature on explanation algorithms, it often remains unclear what precisely these algorithms are for and how they should be used. In this position paper, we argue for a novel and pragmatic perspective: Explainable machine learning needs to recognize its parallels with applied statistics. Concretely, explanations are statistics of high-dimensional functions, and we should think about them analogously to traditional statistical quantities. Among others, this implies that we must think carefully about the matter of interpretation, or how the explanations relate to intuitive questions that humans have about the world. The fact that this is scarcely being discussed in research papers is one of the main drawbacks of the current literature. Luckily, the analogy between explainable machine learning and applied statistics suggests fruitful ways for how research practices can be improved.
Authors: Egor Zverev, Sahar Abdelnabi, Soroush Tabesh, Mario Fritz, Christoph H. Lampert
Abstract: Instruction-tuned Large Language Models (LLMs) show impressive results in numerous practical applications, but they lack essential safety features that are common in other areas of computer science, particularly an explicit separation of instructions and data. This makes them vulnerable to manipulations such as indirect prompt injections and generally unsuitable for safety-critical tasks. Surprisingly, there is currently no established definition or benchmark to quantify this phenomenon. In this work, we close this gap by introducing a formal measure for instruction-data separation and an empirical variant that is calculable from a model's outputs. We also present a new dataset, SEP, that allows estimating the measure for real-world models. Our results on various LLMs show that the problem of instruction-data separation is real: all models fail to achieve high separation, and canonical mitigation techniques, such as prompt engineering and fine-tuning, either fail to substantially improve separation or reduce model utility. The source code and SEP dataset are openly accessible at https://github.com/egozverev/Shold-It-Be-Executed-Or-Processed.
URLs: https://github.com/egozverev/Shold-It-Be-Executed-Or-Processed.
Authors: Alexandra Bazarova, Maria Kovaleva, Ilya Kuleshov, Evgenia Romanenkova, Alexander Stepikin, Alexandr Yugay, Dzhambulat Mollaev, Ivan Kireev, Andrey Savchenko, Alexey Zaytsev
Abstract: In today's world, banks use artificial intelligence to optimize diverse business processes, aiming to improve customer experience. Most of the customer-related tasks can be categorized into two groups: 1) local ones, which focus on a client's current state, such as transaction forecasting, and 2) global ones, which consider the general customer behaviour, e.g., predicting successful loan repayment. Unfortunately, maintaining separate models for each task is costly. Therefore, to better facilitate information management, we compared eight state-of-the-art unsupervised methods on 11 tasks in search for a one-size-fits-all solution. Contrastive self-supervised learning methods were demonstrated to excel at global problems, while generative techniques were superior at local tasks. We also introduced a novel approach, which enriches the client's representation by incorporating external information gathered from other clients. Our method outperforms classical models, boosting accuracy by up to 20\%.
Authors: Shi Yin, Xinyang Pan, Fengyan Wang, Lixin He
Abstract: We propose a framework to combine strong non-linear expressiveness with strict SO(3)-equivariance in prediction of the electronic-structure Hamiltonian, by exploring the mathematical relationships between SO(3)-invariant and SO(3)-equivariant quantities and their representations. The proposed framework, called TraceGrad, first constructs theoretical SO(3)-invariant trace quantities derived from the Hamiltonian targets, and use these invariant quantities as supervisory labels to guide the learning of high-quality SO(3)-invariant features. Given that SO(3)-invariance is preserved under non-linear operations, the learning of invariant features can extensively utilize non-linear mappings, thereby fully capturing the non-linear patterns inherent in physical systems. Building on this, we propose a gradient-based mechanism to induce SO(3)-equivariant encodings of various degrees from the learned SO(3)-invariant features. This mechanism can incorporate powerful non-linear expressive capabilities into SO(3)-equivariant features with consistency of physical dimensions to the regression targets, while theoretically preserving equivariant properties, establishing a strong foundation for predicting Hamiltonian. Our method achieves state-of-the-art performance in prediction accuracy across eight challenging benchmark databases on Hamiltonian prediction. Experimental results demonstrate that this approach not only improves the accuracy of Hamiltonian prediction but also significantly enhances the prediction for downstream physical quantities, and also markedly improves the acceleration performance for the traditional Density Functional Theory algorithms.
Authors: Sho Sonoda, Yuka Hashimoto, Isao Ishikawa, Masahiro Ikeda
Abstract: We present a constructive universal approximation theorem for learning machines equipped with joint-group-equivariant feature maps, called the joint-equivariant machines, based on the group representation theory. "Constructive" here indicates that the distribution of parameters is given in a closed-form expression known as the ridgelet transform. Joint-group-equivariance encompasses a broad class of feature maps that generalize classical group-equivariance. Particularly, fully-connected networks are not group-equivariant but are joint-group-equivariant. Our main theorem also unifies the universal approximation theorems for both shallow and deep networks. Until this study, the universality of deep networks has been shown in a different manner from the universality of shallow networks, but our results discuss them on common ground. Now we can understand the approximation schemes of various learning machines in a unified manner. As applications, we show the constructive universal approximation properties of four examples: depth-$n$ joint-equivariant machine, depth-$n$ fully-connected network, depth-$n$ group-convolutional network, and a new depth-$2$ network with quadratic forms whose universality has not been known.
Authors: Jiwan Seo, Joonhyuk Kang
Abstract: Learning discrete representations with vector quantization (VQ) has emerged as a powerful approach in various generative models. However, most VQ-based models rely on a single, fixed-rate codebook, requiring extensive retraining for new bitrates or efficiency requirements. We introduce Rate-Adaptive Quantization (RAQ), a multi-rate codebook adaptation framework for VQ-based generative models. RAQ applies a data-driven approach to generate variable-rate codebooks from a single baseline VQ model, enabling flexible tradeoffs between compression and reconstruction fidelity. Additionally, we provide a simple clustering-based procedure for pre-trained VQ models, offering an alternative when retraining is infeasible. Our experiments show that RAQ performs effectively across multiple rates, often outperforming conventional fixed-rate VQ baselines. By enabling a single system to seamlessly handle diverse bitrate requirements, RAQ extends the adaptability of VQ-based generative models and broadens their applicability to data compression, reconstruction, and generation tasks.
Authors: Minheng Xiao, Xian Yu, Lei Ying
Abstract: Risk-sensitive reinforcement learning (RL) is crucial for maintaining reliable performance in high-stakes applications. While traditional RL methods aim to learn a point estimate of the random cumulative cost, distributional RL (DRL) seeks to estimate the entire distribution of it, which leads to a unified framework for handling different risk measures. However, developing policy gradient methods for risk-sensitive DRL is inherently more complex as it involves finding the gradient of a probability measure. This paper introduces a new policy gradient method for risk-sensitive DRL with general coherent risk measures, where we provide an analytical form of the probability measure's gradient for any distribution. For practical use, we design a categorical distributional policy gradient algorithm (CDPG) that approximates any distribution by a categorical family supported on some fixed points. We further provide a finite-support optimality guarantee and a finite-iteration convergence guarantee under inexact policy evaluation and gradient estimation. Through experiments on stochastic Cliffwalk and CartPole environments, we illustrate the benefits of considering a risk-sensitive setting in DRL.
Authors: Raja Marjieh, Sreejan Kumar, Declan Campbell, Liyi Zhang, Gianluca Bencomo, Jake Snell, Thomas L. Griffiths
Abstract: Humans rely on effective representations to learn from few examples and abstract useful information from sensory data. Inducing such representations in machine learning models has been shown to improve their performance on various benchmarks such as few-shot learning and robustness. However, finding effective training procedures to achieve that goal can be challenging as psychologically rich training data such as human similarity judgments are expensive to scale, and Bayesian models of human inductive biases are often intractable for complex, realistic domains. Here, we address this challenge by leveraging a Bayesian notion of generative similarity whereby two data points are considered similar if they are likely to have been sampled from the same distribution. This measure can be applied to complex generative processes, including probabilistic programs. We incorporate generative similarity into a contrastive learning objective to enable learning of embeddings that express human cognitive representations. We demonstrate the utility of our approach by showing that it can be used to capture human-like representations of shape regularity, abstract Euclidean geometric concepts, and semantic hierarchies for natural images.
Authors: Ayman Chaouki, Jesse Read, Albert Bifet
Abstract: Decision Tree (DT) Learning is a fundamental problem in Interpretable Machine Learning, yet it poses a formidable optimisation challenge. Practical algorithms have recently emerged, primarily leveraging Dynamic Programming and Branch & Bound. However, most of these approaches rely on a Depth-First-Search strategy, which is inefficient when searching for DTs at high depths and requires the definition of a maximum depth hyperparameter. Best-First-Search was also employed by other methods to circumvent these issues. The downside of this strategy is its higher memory consumption, as such, it has to be designed in a fully efficient manner that takes full advantage of the problem's structure. We formulate the problem as an AND/OR graph search which we solve with a novel AO*-type algorithm called Branches. We prove both optimality and complexity guarantees for Branches and we show that it is more efficient than the state of the art theoretically and on a variety of experiments. Furthermore, Branches supports non-binary features unlike the other methods, we show that this property can further induce larger gains in computational efficiency.
Authors: Piyushi Manupriya, Pratik Jawanpuria, Karthik S. Gurumoorthy, SakethaNath Jagarlapudi, Bamdev Mishra
Abstract: Unbalanced optimal transport (UOT) has recently gained much attention due to its flexible framework for handling un-normalized measures and its robustness properties. In this work, we explore learning (structured) sparse transport plans in the UOT setting, i.e., transport plans have an upper bound on the number of non-sparse entries in each column (structured sparse pattern) or in the whole plan (general sparse pattern). We propose novel sparsity-constrained UOT formulations building on the recently explored maximum mean discrepancy based UOT. We show that the proposed optimization problem is equivalent to the maximization of a weakly submodular function over a uniform matroid or a partition matroid. We develop efficient gradient-based discrete greedy algorithms and provide the corresponding theoretical guarantees. Empirically, we observe that our proposed greedy algorithms select a diverse support set and we illustrate the efficacy of the proposed approach in various applications.
Authors: Emilia Magnani, Marvin Pf\"ortner, Tobias Weber, Philipp Hennig
Abstract: Neural operators generalize neural networks to learn mappings between function spaces from data. They are commonly used to learn solution operators of parametric partial differential equations (PDEs) or propagators of time-dependent PDEs. However, to make them useful in high-stakes simulation scenarios, their inherent predictive error must be quantified reliably. We introduce LUNO, a novel framework for approximate Bayesian uncertainty quantification in trained neural operators. Our approach leverages model linearization to push (Gaussian) weight-space uncertainty forward to the neural operator's predictions. We show that this can be interpreted as a probabilistic version of the concept of currying from functional programming, yielding a function-valued (Gaussian) random process belief. Our framework provides a practical yet theoretically sound way to apply existing Bayesian deep learning methods such as the linearized Laplace approximation to neural operators. Just as the underlying neural operator, our approach is resolution-agnostic by design. The method adds minimal prediction overhead, can be applied post-hoc without retraining the network, and scales to large models and datasets. We evaluate these aspects in a case study on Fourier neural operators.
Authors: Philipp Misof, Pan Kessel, Jan E. Gerken
Abstract: Little is known about the training dynamics of equivariant neural networks, in particular how it compares to data augmented training of their non-equivariant counterparts. Recently, neural tangent kernels (NTKs) have emerged as a powerful tool to analytically study the training dynamics of wide neural networks. In this work, we take an important step towards a theoretical understanding of training dynamics of equivariant models by deriving neural tangent kernels for a broad class of equivariant architectures based on group convolutions. As a demonstration of the capabilities of our framework, we show an interesting relationship between data augmentation and group convolutional networks. Specifically, we prove that they share the same expected prediction at all training times and even off-manifold. In this sense, they have the same training dynamics. We demonstrate in numerical experiments that this still holds approximately for finite-width ensembles. By implementing equivariant NTKs for roto-translations in the plane ($G=C_{n}\ltimes\mathbb{R}^{2}$) and 3d rotations ($G=\mathrm{SO}(3)$), we show that equivariant NTKs outperform their non-equivariant counterparts as kernel predictors for histological image classification and quantum mechanical property prediction.
Authors: Tobias Thummerer, Lars Mikelsons
Abstract: During modeling of dynamical systems, often two or more model architectures are combined to obtain a more powerful or efficient model regarding a specific application area. This covers the combination of multiple machine learning architectures, as well as hybrid models, i.e., the combination of physical simulation models and machine learning. In this work, we briefly discuss which types of model are usually combined in dynamical systems modeling and propose a class of models that is capable of expressing mixed algebraic, discrete, and differential equation-based models. Further, we examine different established, as well as new ways of combining these models from the point of view of system theory and highlight two challenges - algebraic loops and local event functions in discontinuous models - that require a special approach. Finally, we propose a new wildcard architecture that is capable of describing arbitrary combinations of models in an easy-to-interpret fashion that can be learned as part of a gradient-based optimization procedure. In a final experiment, different combination architectures between two models are learned, interpreted, and compared using the methodology and software implementation provided.
Authors: Takuya Ito, Luca Cocchi, Tim Klinger, Parikshit Ram, Murray Campbell, Luke Hearne
Abstract: The attention mechanism is central to the transformer's ability to capture complex dependencies between tokens of an input sequence. Key to the successful application of the attention mechanism in transformers is its choice of positional encoding (PE). The PE provides essential information that distinguishes the position and order amongst tokens in a sequence. Most prior investigations of PE effects on generalization were tailored to 1D input sequences, such as those presented in natural language, where adjacent tokens (e.g., words) are highly related. In contrast, many real world tasks involve datasets with highly non-trivial positional arrangements, such as datasets organized in multiple spatial dimensions, or datasets for which ground truth positions are not known, such as in biological data. Here we study the importance of learning accurate PE for problems which rely on a non-trivial arrangement of input tokens. Critically, we find that the choice of initialization of a learnable PE greatly influences its ability to learn accurate PEs that lead to enhanced generalization. We empirically demonstrate our findings in three experiments: 1) A 2D relational reasoning task; 2) A nonlinear stochastic network simulation; 3) A real world 3D neuroscience dataset, applying interpretability analyses to verify the learning of accurate PEs. Overall, we find that a learned PE initialized from a small-norm distribution can 1) uncover interpretable PEs that mirror ground truth positions in multiple dimensions, and 2) lead to improved downstream generalization in empirical evaluations. Importantly, choosing an ill-suited PE can be detrimental to both model interpretability and generalization. Together, our results illustrate the feasibility of learning identifiable and interpretable PEs for enhanced generalization.
Authors: Akhil Agnihotri, Rahul Jain, Deepak Ramachandran, Zheng Wen
Abstract: Reinforcement Learning with Human Feedback (RLHF) is at the core of fine-tuning methods for generative AI models for language and images. Such feedback is often sought as rank or preference feedback from human raters, as opposed to eliciting scores since the latter tends to be noisy. On the other hand, RL theory and algorithms predominantly assume that a reward feedback is available. In particular, approaches for online learning that can be helpful in adaptive data collection via active learning cannot incorporate offline preference data. In this paper, we adopt a finite-armed linear bandit model as a prototypical model of online learning. We consider an offline preference dataset to be available generated by an expert of unknown `competence'. We propose $\mathsf{warmPref-PS}$, a posterior sampling algorithm for online learning that can be warm-started with an offline dataset with noisy preference feedback. We show that by modeling the `competence' of the expert that generated it, we are able to use such a dataset most effectively. We support our claims with novel theoretical analysis of its Bayesian regret, as well as, extensive empirical evaluation of an approximate loss function that optimizes for infinitely many arms, and performs substantially better (25 to 50\% regret reduction) than baselines.
Authors: Ouya Wang, Shenglong Zhou, Geoffrey Ye Li
Abstract: Stochastic gradient descent-based algorithms are widely used for training deep neural networks but often suffer from slow convergence. To address the challenge, we leverage the framework of the alternating direction method of multipliers (ADMM) to develop a novel data-driven algorithm, called batch ADMM (BADM). The fundamental idea of the proposed algorithm is to split the training data into batches, which is further divided into sub-batches where primal and dual variables are updated to generate global parameters through aggregation. We evaluate the performance of BADM across various deep learning tasks, including graph modelling, computer vision, image generation, and natural language processing. Extensive numerical experiments demonstrate that BADM achieves faster convergence and superior testing accuracy compared to other state-of-the-art optimizers.
Authors: Nan Lin, Peter Palensky, Pedro P. Vergara
Abstract: High-resolution time series data are crucial for the operation and planning of energy systems such as electrical power systems and heating systems. Such data often cannot be shared due to privacy concerns, necessitating the use of synthetic data. However, high-resolution time series data is difficult to model due to its inherent high dimensionality and complex temporal dependencies. Leveraging the recent development of generative AI, especially diffusion models, we propose EnergyDiff, a universal data generation framework for energy time series data. EnergyDiff builds on state-of-the-art denoising diffusion probabilistic models, utilizing a proposed denoising network dedicated to high-resolution time series data and introducing a novel Marginal Calibration technique. Our extensive experimental results demonstrate that EnergyDiff achieves significant improvement in capturing the temporal dependencies and marginal distributions compared to baselines, particularly at the 1-minute resolution. Additionally, EnergyDiff consistently generates highquality time series data across diverse energy domains, time resolutions, and at both customer and transformer levels with reduced computational need.
Authors: Dominik Straub, Tobias F. Niehues, Jan Peters, Constantin A. Rothkopf
Abstract: Bayesian observer and actor models have provided normative explanations for many behavioral phenomena in perception, sensorimotor control, and other areas of cognitive science and neuroscience. They attribute behavioral variability and biases to interpretable entities such as perceptual and motor uncertainty, prior beliefs, and behavioral costs. However, when extending these models to more naturalistic tasks with continuous actions, solving the Bayesian decision-making problem is often analytically intractable. Inverse decision-making, i.e. performing inference over the parameters of such models given behavioral data, is computationally even more difficult. Therefore, researchers typically constrain their models to easily tractable components, such as Gaussian distributions or quadratic cost functions, or resort to numerical approximations. To overcome these limitations, we amortize the Bayesian actor using a neural network trained on a wide range of parameter settings in an unsupervised fashion. Using the pre-trained neural network enables performing efficient gradient-based Bayesian inference of the Bayesian actor model's parameters. We show on synthetic data that the inferred posterior distributions are in close alignment with those obtained using analytical solutions where they exist. Where no analytical solution is available, we recover posterior distributions close to the ground truth. We then show how our method allows for principled model comparison and how it can be used to disentangle factors that may lead to unidentifiabilities between priors and costs. Finally, we apply our method to empirical data from three sensorimotor tasks and compare model fits with different cost functions to show that it can explain individuals' behavioral patterns.
Authors: Carlo Abate, Filippo Maria Bianchi
Abstract: We propose a novel approach to compute the MAXCUT in attributed graphs, i.e., graphs with features associated with nodes and edges. Our approach works well on any kind of graph topology and can find solutions that jointly optimize the MAXCUT along with other objectives. Based on the obtained MAXCUT partition, we implement a hierarchical graph pooling layer for Graph Neural Networks, which is sparse, trainable end-to-end, and particularly suitable for downstream tasks on heterophilic graphs.
Authors: Siqiao Mu, Diego Klabjan
Abstract: Machine unlearning algorithms aim to efficiently remove data from a model without retraining it from scratch, in order to remove corrupted or outdated data or respect a user's ``right to be forgotten." Certified machine unlearning is a strong theoretical guarantee based on differential privacy that quantifies the extent to which an algorithm erases data from the model weights. In contrast to existing works in certified unlearning for convex or strongly convex loss functions, or nonconvex objectives with limiting assumptions, we propose the first, first-order, black-box (i.e., can be applied to models pretrained with vanilla gradient descent) algorithm for unlearning on general nonconvex loss functions, which unlearns by ``rewinding" to an earlier step during the learning process before performing gradient descent on the loss function of the retained data points. We prove $(\epsilon, \delta)$ certified unlearning and performance guarantees that establish the privacy-utility-complexity tradeoff of our algorithm, and we prove generalization guarantees for nonconvex functions that satisfy the Polyak-Lojasiewicz inequality. Finally, we implement our algorithm under a new experimental framework that more accurately reflects real-world use cases for preserving user privacy.
Authors: Nikhil Vyas, Depen Morwani, Rosie Zhao, Mujin Kwun, Itai Shapira, David Brandfonbrener, Lucas Janson, Sham Kakade
Abstract: There is growing evidence of the effectiveness of Shampoo, a higher-order preconditioning method, over Adam in deep learning optimization tasks. However, Shampoo's drawbacks include additional hyperparameters and computational overhead when compared to Adam, which only updates running averages of first- and second-moment quantities. This work establishes a formal connection between Shampoo (implemented with the 1/2 power) and Adafactor -- a memory-efficient approximation of Adam -- showing that Shampoo is equivalent to running Adafactor in the eigenbasis of Shampoo's preconditioner. This insight leads to the design of a simpler and computationally efficient algorithm: $\textbf{S}$hampo$\textbf{O}$ with $\textbf{A}$dam in the $\textbf{P}$reconditioner's eigenbasis (SOAP). With regards to improving Shampoo's computational efficiency, the most straightforward approach would be to simply compute Shampoo's eigendecomposition less frequently. Unfortunately, as our empirical results show, this leads to performance degradation that worsens with this frequency. SOAP mitigates this degradation by continually updating the running average of the second moment, just as Adam does, but in the current (slowly changing) coordinate basis. Furthermore, since SOAP is equivalent to running Adam in a rotated space, it introduces only one additional hyperparameter (the preconditioning frequency) compared to Adam. We empirically evaluate SOAP on language model pre-training with 360m and 660m sized models. In the large batch regime, SOAP reduces the number of iterations by over 40% and wall clock time by over 35% compared to AdamW, with approximately 20% improvements in both metrics compared to Shampoo. An implementation of SOAP is available at https://github.com/nikhilvyas/SOAP.
Authors: H M Mohaimanul Islam, Huynh Q. N. Vo, Paritosh Ramanan
Abstract: Stochastic optimization problems in large-scale multi-stakeholder networked systems (e.g., power grids and supply chains) rely on data-driven scenarios to encapsulate complex spatiotemporal interdependencies. However, centralized aggregation of stakeholder data is challenging due to the existence of data silos resulting from computational and logistical bottlenecks. In this paper, we present SplitVAEs, a decentralized scenario generation framework that leverages variational autoencoders to generate high-quality scenarios without moving stakeholder data. With the help of experiments on distributed memory systems, we demonstrate the broad applicability of SplitVAEs in a variety of domain areas that are dominated by a large number of stakeholders. Our experiments indicate that SplitVAEs can learn spatial and temporal interdependencies in large-scale networks to generate scenarios that match the joint historical distribution of stakeholder data in a decentralized manner. Our experiments show that SplitVAEs deliver robust performance compared to centralized, state-of-the-art benchmark methods while significantly reducing data transmission costs, leading to a scalable, privacy-enhancing alternative to scenario generation.
Authors: Alberto Archetti, Eugenio Lomurno, Diego Piccinotti, Matteo Matteucci
Abstract: Survival analysis is a statistical framework for modeling time-to-event data. It plays a pivotal role in medicine, reliability engineering, and social science research, where understanding event dynamics even with few data samples is critical. Recent advancements in machine learning, particularly those employing neural networks and decision trees, have introduced sophisticated algorithms for survival modeling. However, many of these methods rely on restrictive assumptions about the underlying event-time distribution, such as proportional hazard, time discretization, or accelerated failure time. In this study, we propose FPBoost, a survival model that combines a weighted sum of fully parametric hazard functions with gradient boosting. Distribution parameters are estimated with decision trees trained by maximizing the full survival likelihood. We show how FPBoost is a universal approximator of hazard functions, offering full event-time modeling flexibility while maintaining interpretability through the use of well-established parametric distributions. We evaluate concordance and calibration of FPBoost across multiple benchmark datasets, showcasing its robustness and versatility as a new tool for survival estimation.
Authors: Keitaro Sakamoto, Issei Sato
Abstract: Attention mechanism is a fundamental component of the transformer model and plays a significant role in its success. However, the theoretical understanding of how attention learns to select tokens is still an emerging area of research. In this work, we study the training dynamics and generalization ability of the attention mechanism under classification problems with label noise. We show that, with the characterization of signal-to-noise ratio (SNR), the token selection of attention mechanism achieves benign overfitting, i.e., maintaining high generalization performance despite fitting label noise. Our work also demonstrates an interesting delayed acquisition of generalization after an initial phase of overfitting. Finally, we provide experiments to support our theoretical analysis using both synthetic and real-world datasets.
Authors: Laixi Shi, Jingchu Gai, Eric Mazumdar, Yuejie Chi, Adam Wierman
Abstract: Standard multi-agent reinforcement learning (MARL) algorithms are vulnerable to sim-to-real gaps. To address this, distributionally robust Markov games (RMGs) have been proposed to enhance robustness in MARL by optimizing the worst-case performance when game dynamics shift within a prescribed uncertainty set. RMGs remains under-explored, from reasonable problem formulation to the development of sample-efficient algorithms. Two notorious and open challenges are the formulation of the uncertainty set and whether the corresponding RMGs can overcome the curse of multiagency, where the sample complexity scales exponentially with the number of agents. In this work, we propose a natural class of RMGs inspired by behavioral economics, where each agent's uncertainty set is shaped by both the environment and the integrated behavior of other agents. We first establish the well-posedness of this class of RMGs by proving the existence of game-theoretic solutions such as robust Nash equilibria and coarse correlated equilibria (CCE). Assuming access to a generative model, we then introduce a sample-efficient algorithm for learning the CCE whose sample complexity scales polynomially with all relevant parameters. To the best of our knowledge, this is the first algorithm to break the curse of multiagency for RMGs, regardless of the uncertainty set formulation.
Authors: Kristian Schwethelm, Johannes Kaiser, Jonas Kuntzer, Mehmet Yigitsoy, Daniel Rueckert, Georgios Kaissis
Abstract: Active learning (AL) is a widely used technique for optimizing data labeling in machine learning by iteratively selecting, labeling, and training on the most informative data. However, its integration with formal privacy-preserving methods, particularly differential privacy (DP), remains largely underexplored. While some works have explored differentially private AL for specialized scenarios like online learning, the fundamental challenge of combining AL with DP in standard learning settings has remained unaddressed, severely limiting AL's applicability in privacy-sensitive domains. This work addresses this gap by introducing differentially private active learning (DP-AL) for standard learning settings. We demonstrate that naively integrating DP-SGD training into AL presents substantial challenges in privacy budget allocation and data utilization. To overcome these challenges, we propose step amplification, which leverages individual sampling probabilities in batch creation to maximize data point participation in training steps, thus optimizing data utilization. Additionally, we investigate the effectiveness of various acquisition functions for data selection under privacy constraints, revealing that many commonly used functions become impractical. Our experiments on vision and natural language processing tasks show that DP-AL can improve performance for specific datasets and model architectures. However, our findings also highlight the limitations of AL in privacy-constrained environments, emphasizing the trade-offs between privacy, model accuracy, and data selection accuracy.
Authors: Quan Nguyen, Nishant A. Mehta, Crist\'obal Guzm\'an
Abstract: The minimax sample complexity of group distributionally robust optimization (GDRO) has been determined up to a $\log(K)$ factor, where $K$ is the number of groups. In this work, we venture beyond the minimax perspective via a novel notion of sparsity that we dub $(\lambda, \beta)$-sparsity. In short, this condition means that at any parameter $\theta$, there is a set of at most $\beta$ groups whose risks at $\theta$ all are at least $\lambda$ larger than the risks of the other groups. To find an $\epsilon$-optimal $\theta$, we show via a novel algorithm and analysis that the $\epsilon$-dependent term in the sample complexity can swap a linear dependence on $K$ for a linear dependence on the potentially much smaller $\beta$. This improvement leverages recent progress in sleeping bandits, showing a fundamental connection between the two-player zero-sum game optimization framework for GDRO and per-action regret bounds in sleeping bandits. We next show an adaptive algorithm which, up to log factors, gets a sample complexity bound that adapts to the best $(\lambda, \beta)$-sparsity condition that holds. We also show how to get a dimension-free semi-adaptive sample complexity bound with a computationally efficient method. Finally, we demonstrate the practicality of the $(\lambda, \beta)$-sparsity condition and the improved sample efficiency of our algorithms on both synthetic and real-life datasets.
Authors: Junwoo Ha, Hyukjae Kwon, Sungsoo Kim, Kisu Lee, Seungjae Park, Ha Young Kim
Abstract: Multivariate time series (MTS) forecasting is vital across various domains but remains challenging due to the need to simultaneously model temporal and inter-variate dependencies. Existing channel-dependent models, where Transformer-based models dominate, process these dependencies separately, limiting their capacity to capture complex interactions such as lead-lag dynamics. To address this issue, we propose TiVaT (Time-variate Transformer), a novel architecture incorporating a single unified module, a Joint-Axis (JA) attention module, that concurrently processes temporal and variate modeling. The JA attention module dynamically selects relevant features to particularly capture asynchronous interactions. In addition, we introduce distance-aware time-variate sampling in the JA attention, a novel mechanism that extracts significant patterns through a learned 2D embedding space while reducing noise. Extensive experiments demonstrate TiVaT's overall performance across diverse datasets, particularly excelling in scenarios with intricate asynchronous dependencies.
Authors: Brahma S. Pavse, Yudong Chen, Qiaomin Xie, Josiah P. Hanna
Abstract: In reinforcement learning, offline value function learning is the procedure of using an offline dataset to estimate the expected discounted return from each state when taking actions according to a fixed target policy. The stability of this procedure, i.e., whether it converges to its fixed-point, critically depends on the representations of the state-action pairs. Poorly learned representations can make value function learning unstable, or even divergent. Therefore, it is critical to stabilize value function learning by explicitly shaping the state-action representations. Recently, the class of bisimulation-based algorithms have shown promise in shaping representations for control. However, it is still unclear if this class of methods can stabilize value function learning. In this work, we investigate this question and answer it affirmatively. We introduce a bisimulation-based algorithm called kernel representations for offline policy evaluation (KROPE). KROPE uses a kernel to shape state-action representations such that state-action pairs that have similar immediate rewards and lead to similar next state-action pairs under the target policy also have similar representations. We show that KROPE: 1) learns stable representations and 2) leads to lower value error than baselines. Our analysis provides new theoretical insight into the stability properties of bisimulation-based methods and suggests that practitioners can use these methods for stable and accurate evaluation of offline reinforcement learning agents.
Authors: Leonardo Ferreira Guilhoto, Paris Perdikaris
Abstract: This paper explores alternative formulations of the Kolmogorov Superposition Theorem (KST) as a foundation for neural network design. The original KST formulation, while mathematically elegant, presents practical challenges due to its limited insight into the structure of inner and outer functions and the large number of unknown variables it introduces. Kolmogorov-Arnold Networks (KANs) leverage KST for function approximation, but they have faced scrutiny due to mixed results compared to traditional multilayer perceptrons (MLPs) and practical limitations imposed by the original KST formulation. To address these issues, we introduce ActNet, a scalable deep learning model that builds on the KST and overcomes many of the drawbacks of Kolmogorov's original formulation. We evaluate ActNet in the context of Physics-Informed Neural Networks (PINNs), a framework well-suited for leveraging KST's strengths in low-dimensional function approximation, particularly for simulating partial differential equations (PDEs). In this challenging setting, where models must learn latent functions without direct measurements, ActNet consistently outperforms KANs across multiple benchmarks and is competitive against the current best MLP-based approaches. These results present ActNet as a promising new direction for KST-based deep learning applications, particularly in scientific computing and PDE simulation tasks.
Authors: Shuangpeng Han, Mengmi Zhang
Abstract: AI models make mistakes when recognizing images-whether in-domain, out-of-domain, or adversarial. Predicting these errors is critical for improving system reliability, reducing costly mistakes, and enabling proactive corrections in real-world applications such as healthcare, finance, and autonomous systems. However, understanding what mistakes AI models make, why they occur, and how to predict them remains an open challenge. Here, we conduct comprehensive empirical evaluations using a "mentor" model-a deep neural network designed to predict another "mentee" model's errors. Our findings show that the mentor excels at learning from a mentee's mistakes on adversarial images with small perturbations and generalizes effectively to predict in-domain and out-of-domain errors of the mentee. Additionally, transformer-based mentor models excel at predicting errors across various mentee architectures. Subsequently, we draw insights from these observations and develop an "oracle" mentor model, dubbed SuperMentor, that can outperform baseline mentors in predicting errors across different error types from the ImageNet-1K dataset. Our framework paves the way for future research on anticipating and correcting AI model behaviors, ultimately increasing trust in AI systems. All code, models, and data will be made publicly available.
Authors: Gustav Wagner Zakarias, Lars Kai Hansen, Zheng-Hua Tan
Abstract: This study presents BiSSL, a novel training framework that utilizes bilevel optimization to enhance the alignment between the pretext pre-training and downstream fine-tuning stages in self-supervised learning. BiSSL formulates the pretext and downstream task objectives as the lower- and upper-level objectives in a bilevel optimization problem and serves as an intermediate training stage within the self-supervised learning pipeline. By explicitly modeling the interdependence of these training stages, BiSSL facilitates enhanced information sharing between them, ultimately leading to a backbone parameter initialization that is better aligned for the downstream task. We propose a versatile training algorithm that alternates between optimizing the two objectives defined in BiSSL, which is applicable to a broad range of pretext and downstream tasks. Using SimCLR and Bootstrap Your Own Latent to pre-train ResNet-50 backbones on the ImageNet dataset, we demonstrate that our proposed framework significantly outperforms the conventional self-supervised learning pipeline on the vast majority of 12 downstream image classification datasets, as well as on object detection. Visualizations of the backbone features provide further evidence that BiSSL improves the downstream task alignment of the backbone features prior to fine-tuning.
Authors: Yihong Dong, Ge Li, Yongding Tao, Xue Jiang, Kechi Zhang, Jia Li, Jinliang Deng, Jing Su, Jun Zhang, Jingjing Xu
Abstract: Despite the remarkable successes of general-purpose neural networks, such as MLPs and Transformers, we find that they exhibit notable shortcomings in modeling and reasoning about periodic phenomena, achieving only marginal performance within the training domain and failing to generalize effectively to out-of-domain (OOD) scenarios. Periodicity is ubiquitous throughout nature and science. Therefore, neural networks should be equipped with the essential ability to model and handle periodicity. In this work, we propose FAN, a novel general-purpose neural network that offers broad applicability similar to MLP while effectively addressing periodicity modeling challenges. Periodicity is naturally integrated into FAN's structure and computational processes by introducing the Fourier Principle. Unlike existing Fourier-based networks, which possess particular periodicity modeling abilities but are typically designed for specific tasks, our approach maintains the general-purpose modeling capability. Therefore, FAN can seamlessly replace MLP in various model architectures with fewer parameters and FLOPs. Through extensive experiments, we demonstrate the superiority of FAN in periodicity modeling tasks and the effectiveness and generalizability of FAN across a range of real-world tasks, e.g., symbolic formula representation, time series forecasting, language modeling, and image recognition.
Authors: Enyi Jiang, David S. Cheung, Gagandeep Singh
Abstract: Existing certified training methods can only train models to be robust against a certain perturbation type (e.g. $l_\infty$ or $l_2$). However, an $l_\infty$ certifiably robust model may not be certifiably robust against $l_2$ perturbation (and vice versa) and also has low robustness against other perturbations (e.g. geometric and patch transformation). By constructing a theoretical framework to analyze and mitigate the tradeoff, we propose the first multi-norm certified training framework \textbf{CURE}, consisting of several multi-norm certified training methods, to attain better \emph{union robustness} when training from scratch or fine-tuning a pre-trained certified model. Inspired by our theoretical findings, we devise bound alignment and connect natural training with certified training for better union robustness. Compared with SOTA-certified training, \textbf{CURE} improves union robustness to $32.0\%$ on MNIST, $25.8\%$ on CIFAR-10, and $10.6\%$ on TinyImagenet across different epsilon values. It leads to better generalization on a diverse set of challenging unseen geometric and patch perturbations to $6.8\%$ and $16.0\%$ on CIFAR-10. Overall, our contributions pave a path towards \textit{universal certified robustness}.
Authors: David X. Wu, Anant Sahai
Abstract: The classic teacher-student model in machine learning posits that a strong teacher supervises a weak student to improve the student's capabilities. We instead consider the inverted situation, where a weak teacher supervises a strong student with imperfect pseudolabels. This paradigm was recently brought forth by Burns et al.'23 and termed \emph{weak-to-strong generalization}. We theoretically investigate weak-to-strong generalization for binary and multilabel classification in a stylized overparameterized spiked covariance model with Gaussian covariates where the weak teacher's pseudolabels are asymptotically like random guessing. Under these assumptions, we provably identify two asymptotic phases of the strong student's generalization after weak supervision: (1) successful generalization and (2) random guessing. Our techniques should eventually extend to weak-to-strong multiclass classification. Towards doing so, we prove a tight lower tail inequality for the maximum of correlated Gaussians, which may be of independent interest. Understanding the multilabel setting reinforces the value of using logits for weak supervision when they are available.
Authors: Sevvandi Kandanaarachchi, Conrad Sanderson, Rob J. Hyndman
Abstract: Detecting anomalies in a temporal sequence of graphs can be applied is areas such as the detection of accidents in transport networks and cyber attacks in computer networks. Existing methods for detecting abnormal graphs can suffer from multiple limitations, such as high false positive rates as well as difficulties with handling variable-sized graphs and non-trivial temporal dynamics. To address this, we propose a technique where temporal dependencies are explicitly modelled via time series analysis of a large set of pertinent graph features, followed by using residuals to remove the dependencies. Extreme Value Theory is then used to robustly model and classify any remaining extremes, aiming to produce low false positives rates. Comparative evaluations on a multitude of graph instances show that the proposed approach obtains considerably better accuracy than TensorSplat and Laplacian Anomaly Detection.
Authors: Yuhao Mao, Yani Zhang, Martin Vechev
Abstract: To provide robustness guarantees, neural network certification methods heavily rely on convex relaxations. The imprecision of these convex relaxations, however, is a major obstacle: even the most precise single-neuron relaxation is incomplete for general ReLU networks, a phenomenon referred to as the single-neuron convex barrier. While heuristic instantiations of multi-neuron relaxations have been proposed to circumvent this barrier in practice, their theoretical properties remain largely unknown. In this work, we conduct the first rigorous study of the expressiveness of multi-neuron relaxations. We first show that the ``$\max$'' function in $\mathbb{R}^d$ can be encoded by a ReLU network and exactly bounded by a multi-neuron relaxation, which is impossible for any single-neuron relaxation. Further, we prove that multi-neuron relaxations can be turned into complete verifiers by semantic-preserving structural transformations or by input space partitioning that enjoys improved worst-case partition complexity. We also show that without these augmentations, the completeness guarantee can no longer be obtained, and the relaxation error of every multi-neuron relaxation can be unbounded. To the best of our knowledge, this is the first work to provide an extensive characterization of multi-neuron relaxations and their expressiveness in neural network certification.
Authors: Wei Yao, Zeliang Zhang, Huayi Tang, Yong Liu
Abstract: Model ensemble adversarial attack has become a powerful method for generating transferable adversarial examples that can target even unknown models, but its theoretical foundation remains underexplored. To address this gap, we provide early theoretical insights that serve as a roadmap for advancing model ensemble adversarial attack. We first define transferability error to measure the error in adversarial transferability, alongside concepts of diversity and empirical model ensemble Rademacher complexity. We then decompose the transferability error into vulnerability, diversity, and a constant, which rigidly explains the origin of transferability error in model ensemble attack: the vulnerability of an adversarial example to ensemble components, and the diversity of ensemble components. Furthermore, we apply the latest mathematical tools in information theory to bound the transferability error using complexity and generalization terms, contributing to three practical guidelines for reducing transferability error: (1) incorporating more surrogate models, (2) increasing their diversity, and (3) reducing their complexity in cases of overfitting. Finally, extensive experiments with 54 models validate our theoretical framework, representing a significant step forward in understanding transferable model ensemble adversarial attacks.
Authors: Chenhao Sun, Yuhao Mao, Mark Niklas M\"uller, Martin Vechev
Abstract: Randomized smoothing is a popular approach for providing certified robustness guarantees against adversarial attacks, and has become an active area of research. Over the past years, the average certified radius (ACR) has emerged as the most important metric for comparing methods and tracking progress in the field. However, in this work, for the first time we show that ACR is a poor metric for evaluating robustness guarantees provided by randomized smoothing. We theoretically prove not only that a trivial classifier can have arbitrarily large ACR, but also that ACR is much more sensitive to improvements on easy samples than on hard ones. Empirically, we confirm that existing training strategies, though improving ACR with different approaches, reduce the model's robustness on hard samples consistently. To strengthen our conclusion, we propose strategies, including explicitly discarding hard samples, reweighting the dataset with approximate certified radius, and extreme optimization for easy samples, to achieve state-of-the-art ACR, without training for robustness on the full data distribution. Overall, our results suggest that ACR has introduced a strong undesired bias to the field, and its application should be discontinued when evaluating randomized smoothing.
Authors: Michael Lan, Philip Torr, Austin Meek, Ashkan Khakzar, David Krueger, Fazl Barez
Abstract: We investigate feature universality in large language models (LLMs), a research field that aims to understand how different models similarly represent concepts in the latent spaces of their intermediate layers. Demonstrating feature universality allows discoveries about latent representations to generalize across several models. However, comparing features across LLMs is challenging due to polysemanticity, in which individual neurons often correspond to multiple features rather than distinct ones, making it difficult to disentangle and match features across different models. To address this issue, we employ a method known as dictionary learning by using sparse autoencoders (SAEs) to transform LLM activations into more interpretable spaces spanned by neurons corresponding to individual features. After matching feature neurons across models via activation correlation, we apply representational space similarity metrics on SAE feature spaces across different LLMs. Our experiments reveal significant similarities in SAE feature spaces across various LLMs, providing new evidence for feature universality.
Authors: \"Ozg\"un Turgut, Philip M\"uller, Martin J. Menten, Daniel Rueckert
Abstract: Recent breakthroughs in natural language processing and computer vision, driven by efficient pre-training on large datasets, have enabled foundation models to excel on a wide range of tasks. However, this potential has not yet been fully realised in time series analysis, as existing methods fail to address the heterogeneity in large time series corpora. Prevalent in domains ranging from medicine to finance, time series vary substantially in characteristics such as variate count, inter-variate relationships, temporal patterns, and sampling frequency. To address this, we introduce a novel pre-training paradigm specifically designed to handle time series heterogeneity. We propose a tokeniser with learnable domain signatures, a dual masking strategy, and a normalised cross-correlation loss, enabling our open model for general time series analysis (OTiS) to efficiently learn from large time series corpora. Extensive benchmarking on diverse tasks, such as classification, regression, and forecasting, demonstrates that OTiS outperforms state-of-the-art baselines. Our code and pre-trained weights are available at https://github.com/oetu/otis.
Authors: Jacob Beck, Shikha Surana, Manus McAuliffe, Oliver Bent, Thomas D. Barrett, Juan Jose Garau Luis, Paul Duckworth
Abstract: Predicting the biophysical and functional properties of proteins is essential for in silico protein design. Machine learning has emerged as a promising technique for such prediction tasks. However, the relative scarcity of in vitro annotations means that these models often have little, or no, specific data on the desired fitness prediction task. As a result of limited data, protein language models (PLMs) are typically trained on general protein sequence modeling tasks, and then fine-tuned, or applied zero-shot, to protein fitness prediction. When no task data is available, the models make strong assumptions about the correlation between the protein sequence likelihood and fitness scores. In contrast, we propose meta-learning over a distribution of standard fitness prediction tasks, and demonstrate positive transfer to unseen fitness prediction tasks. Our method, called Metalic (Meta-Learning In-Context), uses in-context learning and fine-tuning, when data is available, to adapt to new tasks. Crucially, fine-tuning enables considerable generalization, even though it is not accounted for during meta-training. Our fine-tuned models achieve strong results with 18 times fewer parameters than state-of-the-art models. Moreover, our method sets a new state-of-the-art in low-data settings on ProteinGym, an established fitness-prediction benchmark. Due to data scarcity, we believe meta-learning will play a pivotal role in advancing protein engineering.
Authors: Abhijnan Nath, Changsoo Jung, Ethan Seefried, Nikhil Krishnaswamy
Abstract: Traditional RLHF-based LLM alignment methods explicitly maximize the expected rewards from a separate reward model. More recent supervised alignment methods like Direct Preference Optimization (DPO) circumvent this phase to avoid problems including model drift and reward overfitting. Although popular due to its simplicity, DPO and similar direct alignment methods which rely heavily on the Bradley-Terry-based pairwise preference formulation can still lead to degenerate policies when challenged by non-deterministic or noisy preference labels, for example human scoring of two candidate outputs with low confidence. This paper introduces DRDO (Direct Reward Distillation and policy-Optimization), which simultaneously models rewards and preferences to avoid such degeneracy. DRDO directly mimics rewards assigned by an oracle while learning human preferences with a novel preference likelihood formulation. Results on the Ultrafeedback and TL;DR datasets demonstrate that DRDO-trained policies surpass methods such as DPO and e-DPO in terms of expected rewards and are more robust, on average, to noisy preference signals as well as out-of-distribution (OOD) settings.
Authors: Wenlong Wang, Ivana Dusparic, Yucheng Shi, Ke Zhang, Vinny Cahill
Abstract: Model-based reinforcement learning (RL) offers a solution to the data inefficiency that plagues most model-free RL algorithms. However, learning a robust world model often demands complex and deep architectures, which are expensive to compute and train. Within the world model, dynamics models are particularly crucial for accurate predictions, and various dynamics-model architectures have been explored, each with its own set of challenges. Currently, recurrent neural network (RNN) based world models face issues such as vanishing gradients and difficulty in capturing long-term dependencies effectively. In contrast, use of transformers suffers from the well-known issues of self-attention mechanisms, where both memory and computational complexity scale as $O(n^2)$, with $n$ representing the sequence length. To address these challenges we propose a state space model (SSM) based world model, specifically based on Mamba, that achieves $O(n)$ memory and computational complexity while effectively capturing long-term dependencies and facilitating the use of longer training sequences efficiently. We also introduce a novel sampling method to mitigate the suboptimality caused by an incorrect world model in the early stages of training, combining it with the aforementioned technique to achieve a normalised score comparable to other state-of-the-art model-based RL algorithms using only a 7 million trainable parameter world model. This model is accessible and can be trained on an off-the-shelf laptop. Our code is available at https://github.com/realwenlongwang/Drama.git
Authors: Ge Li, Dong Tian, Hongyi Zhou, Xinkai Jiang, Rudolf Lioutikov, Gerhard Neumann
Abstract: This work introduces Transformer-based Off-Policy Episodic Reinforcement Learning (TOP-ERL), a novel algorithm that enables off-policy updates in the ERL framework. In ERL, policies predict entire action trajectories over multiple time steps instead of single actions at every time step. These trajectories are typically parameterized by trajectory generators such as Movement Primitives (MP), allowing for smooth and efficient exploration over long horizons while capturing high-level temporal correlations. However, ERL methods are often constrained to on-policy frameworks due to the difficulty of evaluating state-action values for entire action sequences, limiting their sample efficiency and preventing the use of more efficient off-policy architectures. TOP-ERL addresses this shortcoming by segmenting long action sequences and estimating the state-action values for each segment using a transformer-based critic architecture alongside an n-step return estimation. These contributions result in efficient and stable training that is reflected in the empirical results conducted on sophisticated robot learning environments. TOP-ERL significantly outperforms state-of-the-art RL methods. Thorough ablation studies additionally show the impact of key design choices on the model performance.
Authors: Guozhi Liu, Weiwei Lin, Tiansheng Huang, Ruichao Mo, Qi Mu, Li Shen
Abstract: Harmful fine-tuning attack poses a serious threat to the online fine-tuning service. Vaccine, a recent alignment-stage defense, applies uniform perturbation to all layers of embedding to make the model robust to the simulated embedding drift. However, applying layer-wise uniform perturbation may lead to excess perturbations for some particular safety-irrelevant layers, resulting in defense performance degradation and unnecessary memory consumption. To address this limitation, we propose Targeted Vaccine (T-Vaccine), a memory-efficient safety alignment method that applies perturbation to only selected layers of the model. T-Vaccine follows two core steps: First, it uses gradient norm as a statistical metric to identify the safety-critical layers. Second, instead of applying uniform perturbation across all layers, T-Vaccine only applies perturbation to the safety-critical layers while keeping other layers frozen during training. Results show that T-Vaccine outperforms Vaccine in terms of both defense effectiveness and resource efficiency. Comparison with other defense baselines, e.g., RepNoise and TAR also demonstrate the superiority of T-Vaccine. Notably, T-Vaccine is the first defense that can address harmful fine-tuning issues for a 7B pre-trained models trained on consumer GPUs with limited memory (e.g., RTX 4090). Our code is available at https://github.com/Lslland/T-Vaccine.
Authors: Hanbo Huang, Yihan Li, Bowen Jiang, Lin Liu, Bo Jiang, Ruoyu Sun, Zhuotao Liu, Shiyu Liang
Abstract: Current LLM customization typically relies on two deployment strategies: closed-source APIs, which require users to upload private data to external servers, and open-weight models, which allow local fine-tuning but pose misuse risks. In this position paper, we argue that (1) deploying closed-source LLMs within user-controlled infrastructure (\textit{on-premises deployment}) enhances data privacy and mitigates misuse risks, and (2) a well-designed on-premises deployment must ensure model confidentiality -- by preventing model theft -- and offer privacy-preserving customization. Prior research on small models has explored securing only the output layer within hardware-secured devices to balance confidentiality and customization efficiency. However, we show that this approach is insufficient for defending large-scale LLMs against distillation attacks. We therefore introduce a {semi-open deployment framework} that secures only a few, carefully chosen layers, achieving distillation resistance comparable to fully secured models while preserving fine-tuning flexibility. Through extensive experiments, we show that securing bottom layers significantly reduces functional extraction risks. Our findings demonstrate that privacy and confidentiality can coexist, paving the way for secure on-premises AI deployment that balances usability and protection.
Authors: Eric Elmoznino, Tom Marty, Tejas Kasetty, Leo Gagnon, Sarthak Mittal, Mahan Fathi, Dhanya Sridhar, Guillaume Lajoie
Abstract: A central goal of machine learning is generalization. While the No Free Lunch Theorem states that we cannot obtain theoretical guarantees for generalization without further assumptions, in practice we observe that simple models which explain the training data generalize best: a principle called Occam's razor. Despite the need for simple models, most current approaches in machine learning only minimize the training error, and at best indirectly promote simplicity through regularization or architecture design. Here, we draw a connection between Occam's razor and in-context learning: an emergent ability of certain sequence models like Transformers to learn at inference time from past observations in a sequence. In particular, we show that the next-token prediction loss used to train in-context learners is directly equivalent to a data compression technique called prequential coding, and that minimizing this loss amounts to jointly minimizing both the training error and the complexity of the model that was implicitly learned from context. Our theory and the empirical experiments we use to support it not only provide a normative account of in-context learning, but also elucidate the shortcomings of current in-context learning methods, suggesting ways in which they can be improved. We make our code available at https://github.com/3rdCore/PrequentialCode.
Authors: Anjiang Wei, Allen Nie, Thiago S. F. X. Teixeira, Rohan Yadav, Wonchan Lee, Ke Wang, Alex Aiken
Abstract: Modern scientific discovery increasingly relies on high-performance computing for complex modeling and simulation. A key challenge in improving parallel program performance is efficiently mapping tasks to processors and data to memory, a process dictated by intricate, low-level system code known as mappers. Developing high-performance mappers demands days of manual tuning, posing a significant barrier for domain scientists without systems expertise. We introduce a framework that automates mapper development with generative optimization, leveraging richer feedback beyond scalar performance metrics. Our approach features the Agent-System Interface, which includes a Domain-Specific Language (DSL) to abstract away low-level complexity of system code and define a structured search space, as well as AutoGuide, a mechanism that interprets raw execution output into actionable feedback. Unlike traditional reinforcement learning methods such as OpenTuner, which rely solely on scalar feedback, our method finds superior mappers in far fewer iterations. With just 10 iterations, it outperforms OpenTuner even after 1000 iterations, achieving 3.8X faster performance. Our approach finds mappers that surpass expert-written mappers by up to 1.34X speedup across nine benchmarks while reducing tuning time from days to minutes.
Authors: Pau Ferrer-Cid, Jose M. Barcelo-Ordinas, Jorge Garcia-Vidal
Abstract: The development of Internet of Things (IoT) technologies has led to the widespread adoption of monitoring networks for a wide variety of applications, such as smart cities, environmental monitoring, and precision agriculture. A major research focus in recent years has been the development of graph-based techniques to improve the quality of data from sensor networks, a key aspect for the use of sensed data in decision-making processes, digital twins, and other applications. Emphasis has been placed on the development of machine learning and signal processing techniques over graphs, taking advantage of the benefits provided by the use of structured data through a graph topology. Many technologies such as the graph signal processing (GSP) or the successful graph neural networks (GNNs) have been used for data quality enhancement tasks. In this survey, we focus on graph-based models for data quality control in monitoring sensor networks. Furthermore, we delve into the technical details that are commonly leveraged for providing powerful graph-based solutions for data quality tasks in sensor networks, including missing value imputation, outlier detection, or virtual sensing. To conclude, we have identified future trends and challenges such as graph-based models for digital twins or model transferability and generalization.
Authors: Minseon Gwak, Seongrok Moon, Joohwan Ko, PooGyeon Park
Abstract: Due to the lack of state dimension optimization methods, deep state space models (SSMs) have sacrificed model capacity, training search space, or stability to alleviate computational costs caused by high state dimensions. In this work, we provide a structured pruning method for SSMs, Layer-Adaptive STate pruning (LAST), which reduces the state dimension of each layer in minimizing model-level output energy loss by extending modal truncation for a single system. LAST scores are evaluated using the $\mathcal{H}_{\infty}$ norms of subsystems and layer-wise energy normalization. The scores serve as global pruning criteria, enabling cross-layer comparison of states and layer-adaptive pruning. Across various sequence benchmarks, LAST optimizes previous SSMs, revealing the redundancy and compressibility of their state spaces. Notably, we demonstrate that, on average, pruning 33% of states still maintains performance with 0.52% accuracy loss in multi-input multi-output SSMs without retraining. Code is available at https://github.com/msgwak/LAST.
Authors: Arnab Bhattacharyya, Davin Choo, Philips George John, Themis Gouleakis
Abstract: We revisit the problem of distribution learning within the framework of learning-augmented algorithms. In this setting, we explore the scenario where a probability distribution is provided as potentially inaccurate advice on the true, unknown distribution. Our objective is to develop learning algorithms whose sample complexity decreases as the quality of the advice improves, thereby surpassing standard learning lower bounds when the advice is sufficiently accurate. Specifically, we demonstrate that this outcome is achievable for the problem of learning a multivariate Gaussian distribution $N(\boldsymbol{\mu}, \boldsymbol{\Sigma})$ in the PAC learning setting. Classically, in the advice-free setting, $\tilde{\Theta}(d^2/\varepsilon^2)$ samples are sufficient and worst case necessary to learn $d$-dimensional Gaussians up to TV distance $\varepsilon$ with constant probability. When we are additionally given a parameter $\tilde{\boldsymbol{\Sigma}}$ as advice, we show that $\tilde{O}(d^{2-\beta}/\varepsilon^2)$ samples suffices whenever $\| \tilde{\boldsymbol{\Sigma}}^{-1/2} \boldsymbol{\Sigma} \tilde{\boldsymbol{\Sigma}}^{-1/2} - \boldsymbol{I_d} \|_1 \leq \varepsilon d^{1-\beta}$ (where $\|\cdot\|_1$ denotes the entrywise $\ell_1$ norm) for any $\beta > 0$, yielding a polynomial improvement over the advice-free setting.
Authors: Kaizhao Liang, Lizhang Chen, Bo Liu, Qiang Liu
Abstract: AdamW has been the default optimizer for transformer pretraining. For many years, our community searched for faster and more stable optimizers with only constrained positive outcomes. In this work, we propose a single-line modification in Pytorch to any momentum-based optimizer, which we rename cautious optimizer, e.g. C-AdamW and C-Lion. Our theoretical result shows that this modification preserves Adam's Hamiltonian function and it does not break the convergence guarantee under the Lyapunov analysis. In addition, a whole new family of optimizers is revealed by our theoretical insight. Among them, we pick the simplest one for empirical experiments, showing not only speed-up on Llama and MAE pretraining up to $1.47$ times, but also better results in LLM post-training tasks. Code is available at https://github.com/kyleliang919/C-Optim.
Authors: Alexander Capstick, Rahul G. Krishnan, Payam Barnaghi
Abstract: Large language models (LLMs) acquire a breadth of information across various domains. However, their computational complexity, cost, and lack of transparency often hinder their direct application for predictive tasks where privacy and interpretability are paramount. In fields such as healthcare, biology, and finance, specialised and interpretable linear models still hold considerable value. In such domains, labelled data may be scarce or expensive to obtain. Well-specified prior distributions over model parameters can reduce the sample complexity of learning through Bayesian inference; however, eliciting expert priors can be time-consuming. We therefore introduce AutoElicit to extract knowledge from LLMs and construct priors for predictive models. We show these priors are informative and can be refined using natural language. We perform a careful study contrasting AutoElicit with in-context learning and demonstrate how to perform model selection between the two methods. We find that AutoElicit yields priors that can substantially reduce error over uninformative priors, using fewer labels, and consistently outperform in-context learning. We show that AutoElicit saves over 6 months of labelling effort when building a new predictive model for urinary tract infections from sensor recordings of people living with dementia.
Authors: Fanxu Meng, Pingzhi Tang, Fan jiang, Muhan Zhang
Abstract: Decoder-only models generate tokens autoregressively by caching key/value vectors, but as the cache grows, inference becomes memory-bound. To address this issue, we introduce CLOVER (Cross-Layer Orthogonal Vectors), a novel approach that treats pairs of attention layers as a set of low-rank decompositions. CLOVER applies Singular Value Decomposition (SVD) to the \( Q \)-\( K \) and \( V \)-\( O \) pairs within each attention head. The resulting singular values can either guide pruning or serve as trainable parameters for efficient fine-tuning of all orthogonal vectors. After pruning or fine-tuning, these values are reintegrated into the model without increasing its parameter count. We apply CLOVER to various models, including GPT-2 XL, DeepSeek-V2-Lite, Whisper-Large-v3, Stable Diffusion XL, and LLaMA-3.2-11B-Vision. Our results demonstrate that CLOVER significantly improves pruning efficiency. For instance, the perplexity of pruning 70\% of the \( Q \)-\( K \) pairs in GPT-2 XL is similar to that of pruning just 8\% with vanilla methods. Fine-tuning the singular values further results in a full-rank update, outperforming state-of-the-art methods (LoRA, DoRA, HiRA, and PiSSA) by 7.6\%, 5.5\%, 3.8\%, and 0.7\%, respectively, on eight commonsense tasks for LLaMA-2 7B.
Authors: Wonkyo Choe, Yangfeng Ji, Felix Xiaozhu Lin
Abstract: To deploy LLMs on resource-contained platforms such as mobile robots and smartphones, non-transformers LLMs have achieved major breakthroughs. Recently, a novel RNN-based LLM family, Repentance Weighted Key Value (RWKV) has shown strong computational efficiency; nevertheless, RWKV models still have high parameter counts which limited their deployment. In this paper, we propose a suite of compression techniques, ranging from model architecture optimizations to post-training compression, tailored to the RWKV architecture. Combined, our techniques reduce the memory footprint of RWKV models by 3.4x -- 5x with only negligible degradation in accuracy; compared to transformer LLMs with similar accuracy, our models require 4x less memory footprint.
Authors: Salar Abbaspourazad, Anshuman Mishra, Joseph Futoma, Andrew C. Miller, Ian Shapiro
Abstract: Modern wearable devices can conveniently record various biosignals in the many different environments of daily living, enabling a rich view of individual health. However, not all biosignals are the same: high-fidelity biosignals, such as photoplethysmogram (PPG), contain more physiological information, but require optical sensors with a high power footprint. Alternatively, a lower-fidelity biosignal such as accelerometry has a significantly smaller power footprint and is available in almost any wearable device. While accelerometry is widely used for activity recognition and fitness, it is less explored for health biomarkers and diagnosis. Here, we show that an accelerometry foundation model can predict a wide variety of health targets. To achieve improved performance, we distill representational knowledge from PPG encoders to accelerometery encoders using 20 million minutes of unlabeled data, collected from ~172K participants in the Apple Heart and Movement Study under informed consent. We observe strong cross-modal alignment on unseen data, e.g., 99.2% top-1 accuracy for retrieving PPG embeddings from accelerometry embeddings. We show that distilled accelerometry encoders have significantly more informative representations compared to self-supervised or supervised encoders trained directly on accelerometry data, observed by at least 23%-49% improved performance for predicting heart rate and heart rate variability. We also show that distilled accelerometry encoders are readily predictive of a wide array of downstream health targets, i.e., they are generalist foundation models. We believe accelerometry foundation models for health may unlock new opportunities for developing digital biomarkers from any wearable device.
Authors: Georgios Tertytchny, Georgios L. Stavrinides, Maria K. Michael
Abstract: To address the challenges of imbalanced multi-class datasets typically used for rare event detection in critical cyber-physical systems, we propose an optimal, efficient, and adaptable mixed integer programming (MIP) ensemble weighting scheme. Our approach leverages the diverse capabilities of the classifier ensemble on a granular per class basis, while optimizing the weights of classifier-class pairs using elastic net regularization for improved robustness and generalization. Additionally, it seamlessly and optimally selects a predefined number of classifiers from a given set. We evaluate and compare our MIP-based method against six well-established weighting schemes, using representative datasets and suitable metrics, under various ensemble sizes. The experimental results reveal that MIP outperforms all existing approaches, achieving an improvement in balanced accuracy ranging from 0.99% to 7.31%, with an overall average of 4.53% across all datasets and ensemble sizes. Furthermore, it attains an overall average increase of 4.63%, 4.60%, and 4.61% in macro-averaged precision, recall, and F1-score, respectively, while maintaining computational efficiency.
Authors: Zehong Wang, Zheyuan Zhang, Tianyi Ma, Nitesh V Chawla, Chuxu Zhang, Yanfang Ye
Abstract: Foundation models aim to create general, cross-task, and cross-domain machine learning models by pretraining on large-scale datasets to capture shared patterns or concepts, such as contours, colors, textures, and edges in images, or tokens, words, and sentences in text. However, identifying generalities across graph-structured data remains a significant challenge, as different graph-based tasks necessitate distinct inductive biases, thereby impeding the development of graph foundation models. To address this challenge, we introduce a novel approach for learning cross-task generalities in graphs. Specifically, we propose task-trees as basic learning instances to align task spaces (node, link, graph) on graphs. Then, we conduct a theoretical analysis to examine their stability, transferability, and generalization. Our findings indicate that when a graph neural network (GNN) is pretrained on diverse task-trees using a reconstruction objective, it acquires transferable knowledge, enabling effective adaptation to downstream tasks with an appropriate set of fine-tuning samples. To empirically validate this approach, we develop a pretrained graph model based on task-trees, termed Graph Generality Identifier on Task-Trees (GIT). Extensive experiments demonstrate that a single pretrained GIT model can be effectively adapted to over 30 different graphs across five domains via fine-tuning, in-context learning, or zero-shot learning. Our data and code are available at https://github.com/Zehong-Wang/GIT.
Authors: Arseniy Andreyev, Pierfrancesco Beneventano
Abstract: Recent findings by Cohen et al., 2021, demonstrate that when training neural networks with full-batch gradient descent with a step size of $\eta$, the largest eigenvalue $\lambda_{\max}$ of the full-batch Hessian consistently stabilizes at $\lambda_{\max} = 2/\eta$. These results have significant implications for convergence and generalization. This, however, is not the case of mini-batch stochastic gradient descent (SGD), limiting the broader applicability of its consequences. We show that SGD trains in a different regime we term Edge of Stochastic Stability (EoSS). In this regime, what stabilizes at $2/\eta$ is *Batch Sharpness*: the expected directional curvature of mini-batch Hessians along their corresponding stochastic gradients. As a consequence $\lambda_{\max}$--which is generally smaller than Batch Sharpness--is suppressed, aligning with the long-standing empirical observation that smaller batches and larger step sizes favor flatter minima. We further discuss implications for mathematical modeling of SGD trajectories.
Authors: Yuanchao Xu, Kaidi Shao, Nikos Logothetis, Zhongwei Shen
Abstract: Analyzing long-term behaviors in high-dimensional nonlinear dynamical systems remains challenging, with the Koopman operator framework providing a powerful global linearization approach, though existing methods for approximating its spectral components often suffer from theoretical limitations and reliance on predefined dictionaries. While Residual Dynamic Mode Decomposition (ResDMD) introduced the spectral residual to assess the accuracy of Koopman operator approximation, its only filters precomputed spectra, which prevents it from fully discovering the Koopman operator's complete spectral information (a limitation sometimes referred to as the 'spectral inclusion' problem). We introduce ResKoopNet (Residual-based Koopman-learning Network), a novel method that addresses this limitation by explicitly minimizing the spectral residual to compute Koopman eigenpairs, which can identify a more precise and complete spectrum of the Koopman operator. This approach provides theoretical guarantees while maintaining computational adaptability through a neural network implementation. Experiments on physical and biological systems demonstrate ResKoopNet's superior accuracy in spectral approximation compared to existing methods, particularly for systems with continuous spectra and high dimensional, which makes it as an effective tool for analyzing complex dynamical systems.
Authors: Kunpeng Xu, Lifei Chen, Shengrui Wang
Abstract: In the realm of time series analysis, tackling the phenomenon of concept drift poses a significant challenge. Concept drift -- characterized by the evolving statistical properties of time series data, affects the reliability and accuracy of conventional analysis models. This is particularly evident in co-evolving scenarios where interactions among variables are crucial. This paper presents CORAL, a simple yet effective method that models time series as an evolving ecosystem to learn representations of concept drift. CORAL employs a kernel-induced self-representation learning to generate a representation matrix, encapsulating the inherent dynamics of co-evolving time series. This matrix serves as a key tool for identification and adaptation to concept drift by observing its temporal variations. Furthermore, CORAL effectively identifies prevailing patterns and offers insights into emerging trends through pattern evolution analysis. Our empirical evaluation of CORAL across various datasets demonstrates its effectiveness in handling the complexities of concept drift. This approach introduces a novel perspective in the theoretical domain of co-evolving time series analysis, enhancing adaptability and accuracy in the face of dynamic data environments, and can be easily integrated into most deep learning backbones.
Authors: Satchel Grant
Abstract: When can we say that two neural systems are the same? The answer to this question is goal-dependent, and it is often addressed through correlative methods such as Representational Similarity Analysis (RSA) and Centered Kernel Alignment (CKA). How do we target functionally relevant similarity, and how do we isolate specific causal aspects of the representations? In this work, we introduce Model Alignment Search (MAS), a method for causally exploring distributed representational similarity. The method learns invertible linear transformations that align a subspace between two distributed networks' representations where causal information can be freely interchanged. We first show that the method can be used to transfer values of specific causal variables -- such as the number of items in a counting task -- between networks with different training seeds. We then explore open questions in number cognition by comparing different types of numeric representations in models trained on structurally different tasks. We then explore differences between MAS vs preexisting causal similarity methods, and lastly, we introduce a counterfactual latent auxiliary loss function that helps shape causally relevant alignments even in cases where we do not have causal access to one of the two models for training.
Authors: Hoang-Thang Ta, Duy-Quy Thai, Anh Tran, Grigori Sidorov, Alexander Gelbukh
Abstract: Kolmogorov-Arnold Networks (KANs) represent an innovation in neural network architectures, offering a compelling alternative to Multi-Layer Perceptrons (MLPs) in models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Transformers. By advancing network design, KANs drive groundbreaking research and enable transformative applications across various scientific domains involving neural networks. However, existing KANs often require significantly more parameters in their network layers than MLPs. To address this limitation, this paper introduces PRKANs (Parameter-Reduced Kolmogorov-Arnold Networks), which employ several methods to reduce the parameter count in KAN layers, making them comparable to MLP layers. Experimental results on the MNIST and Fashion-MNIST datasets demonstrate that PRKANs outperform several existing KANs, and their variant with attention mechanisms rivals the performance of MLPs, albeit with slightly longer training times. Furthermore, the study highlights the advantages of Gaussian Radial Basis Functions (GRBFs) and layer normalization in KAN designs. The repository for this work is available at: https://github.com/hoangthangta/All-KAN.
Authors: Yan Chen, Qinxun Bai, Yiteng Zhang, Shi Dong, Maria Dimakopoulou, Qi Sun, Zhengyuan Zhou
Abstract: Designing learning agents that explore efficiently in a complex environment has been widely recognized as a fundamental challenge in reinforcement learning. While a number of works have demonstrated the effectiveness of techniques based on randomized value functions on a single agent, it remains unclear, from a theoretical point of view, whether injecting randomization can help a society of agents {\it concurently} explore an environment. The theoretical results %that we established in this work tender an affirmative answer to this question. We adapt the concurrent learning framework to \textit{randomized least-squares value iteration} (RLSVI) with \textit{aggregated state representation}. We demonstrate polynomial worst-case regret bounds in both finite- and infinite-horizon environments. In both setups the per-agent regret decreases at an optimal rate of $\Theta\left(\frac{1}{\sqrt{N}}\right)$, highlighting the advantage of concurent learning. Our algorithm exhibits significantly lower space complexity compared to \cite{russo2019worst} and \cite{agrawal2021improved}. We reduce the space complexity by a factor of $K$ while incurring only a $\sqrt{K}$ increase in the worst-case regret bound, compared to \citep{agrawal2021improved,russo2019worst}. Additionally, we conduct numerical experiments to demonstrate our theoretical findings.
Authors: Milad Khademi Nori, Il-Min Kim, Guanghui Wang
Abstract: Federated Class-Incremental Learning (FCIL) refers to a scenario where a dynamically changing number of clients collaboratively learn an ever-increasing number of incoming tasks. FCIL is known to suffer from local forgetting due to class imbalance at each client and global forgetting due to class imbalance across clients. We develop a mathematical framework for FCIL that formulates local and global forgetting. Then, we propose an approach called Hybrid Rehearsal (HR), which utilizes latent exemplars and data-free techniques to address local and global forgetting, respectively. HR employs a customized autoencoder designed for both data classification and the generation of synthetic data. To determine the embeddings of new tasks for all clients in the latent space of the encoder, the server uses the Lennard-Jones Potential formulations. Meanwhile, at the clients, the decoder decodes the stored low-dimensional latent space exemplars back to the high-dimensional input space, used to address local forgetting. To overcome global forgetting, the decoder generates synthetic data. Furthermore, our mathematical framework proves that our proposed approach HR can, in principle, tackle the two local and global forgetting challenges. In practice, extensive experiments demonstrate that while preserving privacy, our proposed approach outperforms the state-of-the-art baselines on multiple FCIL benchmarks with low compute and memory footprints.
Authors: Leandro Soares, Gustavo Venturini, Jos\'e Gomes, Jonathan Efigenio, Pablo Rangel, Pedro Gonzalez, Joel dos Santos, Diego Brand\~ao, Eduardo Bezerra
Abstract: Military personnel and security agents often face significant physical risks during conflict and engagement situations, particularly in urban operations. Ensuring the rapid and accurate communication of incidents involving injuries is crucial for the timely execution of rescue operations. This article presents research conducted under the scope of the Brazilian Navy's ``Soldier of the Future'' project, focusing on the development of a Casualty Detection System to identify injuries that could incapacitate a soldier and lead to severe blood loss. The study specifically addresses the detection of soldier falls, which may indicate critical injuries such as hypovolemic hemorrhagic shock. To generate the publicly available dataset, we used smartwatches and smartphones as wearable devices to collect inertial data from soldiers during various activities, including simulated falls. The data were used to train 1D Convolutional Neural Networks (CNN1D) with the objective of accurately classifying falls that could result from life-threatening injuries. We explored different sensor placements (on the wrists and near the center of mass) and various approaches to using inertial variables, including linear and angular accelerations. The neural network models were optimized using Bayesian techniques to enhance their performance. The best-performing model and its results, discussed in this article, contribute to the advancement of automated systems for monitoring soldier safety and improving response times in engagement scenarios.
Authors: Liam Carroll, Jesse Hoogland, Matthew Farrugia-Roberts, Daniel Murfet
Abstract: Modern deep neural networks display striking examples of rich internal computational structure. Uncovering principles governing the development of such structure is a priority for the science of deep learning. In this paper, we explore the transient ridge phenomenon: when transformers are trained on in-context linear regression tasks with intermediate task diversity, they initially behave like ridge regression before specializing to the tasks in their training distribution. This transition from a general solution to a specialized solution is revealed by joint trajectory principal component analysis. Further, we draw on the theory of Bayesian internal model selection to suggest a general explanation for the phenomena of transient structure in transformers, based on an evolving tradeoff between loss and complexity. We empirically validate this explanation by measuring the model complexity of our transformers as defined by the local learning coefficient.
Authors: Bartosz Cywi\'nski, Kamil Deja
Abstract: Diffusion models, while powerful, can inadvertently generate harmful or undesirable content, raising significant ethical and safety concerns. Recent machine unlearning approaches offer potential solutions but often lack transparency, making it difficult to understand the changes they introduce to the base model. In this work, we introduce SAeUron, a novel method leveraging features learned by sparse autoencoders (SAEs) to remove unwanted concepts in text-to-image diffusion models. First, we demonstrate that SAEs, trained in an unsupervised manner on activations from multiple denoising timesteps of the diffusion model, capture sparse and interpretable features corresponding to specific concepts. Building on this, we propose a feature selection method that enables precise interventions on model activations to block targeted content while preserving overall performance. Evaluation with the competitive UnlearnCanvas benchmark on object and style unlearning highlights SAeUron's state-of-the-art performance. Moreover, we show that with a single SAE, we can remove multiple concepts simultaneously and that in contrast to other methods, SAeUron mitigates the possibility of generating unwanted content, even under adversarial attack. Code and checkpoints are available at: https://github.com/cywinski/SAeUron.
Authors: Yunzhen Yao, Lie He, Michael Gastpar
Abstract: This paper considers the sample-efficiency of preference learning, which models and predicts human choices based on comparative judgments. The minimax optimal estimation rate $\Theta(d/n)$ in traditional estimation theory requires that the number of samples $n$ scales linearly with the dimensionality of the feature space $d$. However, the high dimensionality of the feature space and the high cost of collecting human-annotated data challenge the efficiency of traditional estimation methods. To remedy this, we leverage sparsity in the preference model and establish sharp estimation rates. We show that under the sparse random utility model, where the parameter of the reward function is $k$-sparse, the minimax optimal rate can be reduced to $\Theta(k/n \log(d/k))$. Furthermore, we analyze the $\ell_{1}$-regularized estimator and show that it achieves near-optimal rate under mild assumptions on the Gram matrix. Experiments on synthetic data and LLM alignment data validate our theoretical findings, showing that sparsity-aware methods significantly reduce sample complexity and improve prediction accuracy.
Authors: Alessandro Barro
Abstract: This study focuses on the application of deep geometric models to solve the 3x3x3 Rubik's Cube. We begin by discussing the cube's graph representation and defining distance as the model's optimization objective. The distance approximation task is reformulated as a node classification problem, effectively addressed using Graph Neural Networks (GNNs). After training the model on a random subgraph, the predicted classes are used to construct a heuristic for $A^*$ search. We conclude with experiments comparing our heuristic to that of the DeepCubeA model.
Authors: Spandan Madan, Tomotake Sasaki, Hanspeter Pfister, Tzu-Mao Li, Xavier Boix
Abstract: Neural networks are susceptible to small perturbations in the form of 2D rotations and shifts, image crops, and even changes in object colors. Past works attribute these errors to dataset bias, claiming that models fail on these perturbed samples as they do not belong to the training data distribution. Here, we challenge this claim and present evidence of the widespread existence of perturbed images within the training data distribution, which networks fail to classify. We train models on data sampled from parametric distributions, then search inside this data distribution to find such in-distribution adversarial examples. This is done using our gradient-free evolution strategies (ES) based approach which we call CMA-Search. Despite training with a large-scale (0.5 million images), unbiased dataset of camera and light variations, CMA-Search can find a failure inside the data distribution in over 71% cases by perturbing the camera position. With lighting changes, CMA-Search finds misclassifications in 42% cases. These findings also extend to natural images from ImageNet and Co3D datasets. This phenomenon of in-distribution images presents a highly worrisome problem for artificial intelligence -- they bypass the need for a malicious agent to add engineered noise to induce an adversarial attack. All code, datasets, and demos are available at https://github.com/Spandan-Madan/in_distribution_adversarial_examples.
URLs: https://github.com/Spandan-Madan/in_distribution_adversarial_examples.
Authors: Thomas Nowotny, James P. Turner, James C. Knight
Abstract: Event-based machine learning promises more energy-efficient AI on future neuromorphic hardware. Here, we investigate how the recently discovered Eventprop algorithm for gradient descent on exact gradients in spiking neural networks can be scaled up to challenging keyword recognition benchmarks. We implemented Eventprop in the GPU-enhanced Neural Networks framework and used it for training recurrent spiking neural networks on the Spiking Heidelberg Digits and Spiking Speech Commands datasets. We found that learning depended strongly on the loss function and extended Eventprop to a wider class of loss functions to enable effective training. We then tested a large number of data augmentations and regularisations as well as exploring different network structures; and heterogeneous and trainable timescales. We found that when combined with two specific augmentations, the right regularisation and a delay line input, Eventprop networks with one recurrent layer achieved state-of-the-art performance on Spiking Heidelberg Digits and good accuracy on Spiking Speech Commands. In comparison to a leading surrogate-gradient-based SNN training method, our GeNN Eventprop implementation is 3X faster and uses 4X less memory. This work is a significant step towards a low-power neuromorphic alternative to current machine learning paradigms.
Authors: Amirreza Hashemi, Yuemeng Feng, Arman Rahmim, Hamid Sabet
Abstract: This work investigates use of equivariant neural networks as efficient and high-performance frameworks for image reconstruction and denoising in nuclear medicine. Our work aims to tackle limitations of conventional Convolutional Neural Networks (CNNs), which require significant training. We investigated equivariant networks, aiming to reduce CNN's dependency on specific training sets. Specifically, we implemented and evaluated equivariant spherical CNNs (SCNNs) for 2- and 3-dimensional medical imaging problems. Our results demonstrate superior quality and computational efficiency of SCNNs in both image reconstruction and denoising benchmark problems. Furthermore, we propose a novel approach to employ SCNNs as a complement to conventional image reconstruction tools, enhancing the outcomes while reducing reliance on the training set. Across all cases, we observed significant decrease in computational cost by leveraging the inherent inclusion of equivariant representatives while achieving the same or higher quality of image processing using SCNNs compared to CNNs. Additionally, we explore the potential of SCNNs for broader tomography applications, particularly those requiring rotationally variant representation.
Authors: Gergely Szab\'o, Paolo Bonaiuti, Andrea Ciliberto, Andr\'as Horv\'ath
Abstract: The accurate tracking of live cells using video microscopy recordings remains a challenging task for popular state-of-the-art image processing based object tracking methods. In recent years, several existing and new applications have attempted to integrate deep-learning based frameworks for this task, but most of them still heavily rely on consecutive frame based tracking embedded in their architecture or other premises that hinder generalized learning. To address this issue, we aimed to develop a new deep-learning based tracking method that relies solely on the assumption that cells can be tracked based on their spatio-temporal neighborhood, without restricting it to consecutive frames. The proposed method has the additional benefit that the motion patterns of the cells can be learned completely by the predictor without any prior assumptions, and it has the potential to handle a large number of video frames with heavy artifacts. The efficacy of the proposed method is demonstrated through biologically motivated validation strategies and compared against multiple state-of-the-art cell tracking methods.
Authors: Ryan Sweke, Erik Recio, Sofiene Jerbi, Elies Gil-Fuster, Bryce Fuller, Jens Eisert, Johannes Jakob Meyer
Abstract: Quantum machine learning is arguably one of the most explored applications of near-term quantum devices. Much focus has been put on notions of variational quantum machine learning where parameterized quantum circuits (PQCs) are used as learning models. These PQC models have a rich structure which suggests that they might be amenable to efficient dequantization via random Fourier features (RFF). In this work, we establish necessary and sufficient conditions under which RFF does indeed provide an efficient dequantization of variational quantum machine learning for regression. We build on these insights to make concrete suggestions for PQC architecture design, and to identify structures which are necessary for a regression problem to admit a potential quantum advantage via PQC based optimization.
Authors: Stefan Smeu, Elena Burceanu, Emanuela Haller, Andrei Liviu Nicolicioiu
Abstract: Novelty detection seeks to identify samples deviating from a known distribution, yet data shifts in a multitude of ways, and only a few consist of relevant changes. Aligned with out-of-distribution generalization literature, we advocate for a formal distinction between task-relevant semantic or content changes and irrelevant style changes. This distinction forms the basis for robust novelty detection, emphasizing the identification of semantic changes resilient to style distributional shifts. To this end, we introduce Stylist, a method that utilizes pretrained large-scale model representations to selectively discard environment-biased features. By computing per-feature scores based on feature distribution distances between environments, Stylist effectively eliminates features responsible for spurious correlations, enhancing novelty detection performance. Evaluations on adapted domain generalization datasets and a synthetic dataset demonstrate Stylist's efficacy in improving novelty detection across diverse datasets with stylistic and content shifts. The code is available at https://github.com/bit-ml/Stylist.
Authors: Jeremy Goldwasser, Giles Hooker
Abstract: Feature importance scores are ubiquitous tools for understanding the predictions of machine learning models. However, many popular attribution methods suffer from high instability due to random sampling. Leveraging novel ideas from hypothesis testing, we devise techniques that ensure the most important features are correct with high-probability guarantees. These assess the set of $K$ top-ranked features, as well as the order of its elements. Given a set of local or global importance scores, we demonstrate how to retrospectively verify the stability of the highest ranks. We then introduce two efficient sampling algorithms that identify the $K$ most important features, perhaps in order, with probability exceeding $1-\alpha$. The theoretical justification for these procedures is validated empirically on SHAP and LIME.
Authors: Fabi\'an Aguirre-L\'opez, Silvio Franz, Mauro Pastore
Abstract: Random features models play a distinguished role in the theory of deep learning, describing the behavior of neural networks close to their infinite-width limit. In this work, we present a thorough analysis of the generalization performance of random features models for generic supervised learning problems with Gaussian data. Our approach, built with tools from the statistical mechanics of disordered systems, maps the random features model to an equivalent polynomial model, and allows us to plot average generalization curves as functions of the two main control parameters of the problem: the number of random features $N$ and the size $P$ of the training set, both assumed to scale as powers in the input dimension $D$. Our results extend the case of proportional scaling between $N$, $P$ and $D$. They are in accordance with rigorous bounds known for certain particular learning tasks and are in quantitative agreement with numerical experiments performed over many order of magnitudes of $N$ and $P$. We find good agreement also far from the asymptotic limits where $D\to \infty$ and at least one between $P/D^K$, $N/D^L$ remains finite.
Authors: Yuhao Wu, Franziska Roesner, Tadayoshi Kohno, Ning Zhang, Umar Iqbal
Abstract: Large language models (LLMs) extended as systems, such as ChatGPT, have begun supporting third-party applications. These LLM apps leverage the de facto natural language-based automated execution paradigm of LLMs: that is, apps and their interactions are defined in natural language, provided access to user data, and allowed to freely interact with each other and the system. These LLM app ecosystems resemble the settings of earlier computing platforms, where there was insufficient isolation between apps and the system. Because third-party apps may not be trustworthy, and exacerbated by the imprecision of natural language interfaces, the current designs pose security and privacy risks for users. In this paper, we evaluate whether these issues can be addressed through execution isolation and what that isolation might look like in the context of LLM-based systems, where there are arbitrary natural language-based interactions between system components, between LLM and apps, and between apps. To that end, we propose IsolateGPT, a design architecture that demonstrates the feasibility of execution isolation and provides a blueprint for implementing isolation, in LLM-based systems. We evaluate IsolateGPT against a number of attacks and demonstrate that it protects against many security, privacy, and safety issues that exist in non-isolated LLM-based systems, without any loss of functionality. The performance overhead incurred by IsolateGPT to improve security is under 30% for three-quarters of tested queries.
Authors: JoonHo Lee, Jae Oh Woo, Juree Seok, Parisa Hassanzadeh, Wooseok Jang, JuYoun Son, Sima Didari, Baruch Gutow, Heng Hao, Hankyu Moon, Wenjun Hu, Yeong-Dae Kwon, Taehee Lee, Seungjai Min
Abstract: Assessing response quality to instructions in language models is vital but challenging due to the complexity of human language across different contexts. This complexity often results in ambiguous or inconsistent interpretations, making accurate assessment difficult. To address this issue, we propose a novel Uncertainty-aware Reward Model (URM) that introduces a robust uncertainty estimation for the quality of paired responses based on Bayesian approximation. Trained with preference datasets, our uncertainty-enabled proxy not only scores rewards for responses but also evaluates their inherent uncertainty. Empirical results demonstrate significant benefits of incorporating the proposed proxy into language model training. Our method boosts the instruction following capability of language models by refining data curation for training and improving policy optimization objectives, thereby surpassing existing methods by a large margin on benchmarks such as Vicuna and MT-bench. These findings highlight that our proposed approach substantially advances language model training and paves a new way of harnessing uncertainty within language models.
Authors: Emanuele Loffredo, Mauro Pastore, Simona Cocco, R\'emi Monasson
Abstract: Class imbalance in real-world data poses a common bottleneck for machine learning tasks, since achieving good generalization on under-represented examples is often challenging. Mitigation strategies, such as under or oversampling the data depending on their abundances, are routinely proposed and tested empirically, but how they should adapt to the data statistics remains poorly understood. In this work, we determine exact analytical expressions of the generalization curves in the high-dimensional regime for linear classifiers (Support Vector Machines). We also provide a sharp prediction of the effects of under/oversampling strategies depending on class imbalance, first and second moments of the data, and the metrics of performance considered. We show that mixed strategies involving under and oversampling of data lead to performance improvement. Through numerical experiments, we show the relevance of our theoretical predictions on real datasets, on deeper architectures and with sampling strategies based on unsupervised probabilistic models.
Authors: Nathan Wycoff
Abstract: Regression trees have emerged as a preeminent tool for solving real-world regression problems due to their ability to deal with nonlinearities, interaction effects and sharp discontinuities. In this article, we rather study regression trees applied to well-behaved, differentiable functions, and determine the relationship between node parameters and the local gradient of the function being approximated. We find a simple estimate of the gradient which can be efficiently computed using quantities exposed by popular tree learning libraries. This allows the tools developed in the context of differentiable algorithms, like neural nets and Gaussian processes, to be deployed to tree-based models. To demonstrate this, we study measures of model sensitivity defined in terms of integrals of gradients and demonstrate how to compute them for regression trees using the proposed gradient estimates. Quantitative and qualitative numerical experiments reveal the capability of gradients estimated by regression trees to improve predictive analysis, solve tasks in uncertainty quantification, and provide interpretation of model behavior.
Authors: Daniel Bethell, Simos Gerasimou, Radu Calinescu, Calum Imrie
Abstract: Empowering safe exploration of reinforcement learning (RL) agents during training is a critical challenge towards their deployment in many real-world scenarios. When prior knowledge of the domain or task is unavailable, training RL agents in unknown, \textit{black-box} environments presents an even greater safety risk. We introduce \mbox{ADVICE} (Adaptive Shielding with a Contrastive Autoencoder), a novel post-shielding technique that distinguishes safe and unsafe features of state-action pairs during training, and uses this knowledge to protect the RL agent from executing actions that yield likely hazardous outcomes. Our comprehensive experimental evaluation against state-of-the-art safe RL exploration techniques shows that ADVICE significantly reduces safety violations ($\approx\!\!50\%$) during training, with a competitive outcome reward compared to other techniques.
Authors: Vincent P. Grande, Michael T. Schaub
Abstract: Topological Data Analysis (TDA) allows us to extract powerful topological and higher-order information on the global shape of a data set or point cloud. Tools like Persistent Homology or the Euler Transform give a single complex description of the global structure of the point cloud. However, common machine learning applications like classification require point-level information and features to be available. In this paper, we bridge this gap and propose a novel method to extract node-level topological features from complex point clouds using discrete variants of concepts from algebraic topology and differential geometry. We verify the effectiveness of these topological point features (TOPF) on both synthetic and real-world data and study their robustness under noise and heterogeneous sampling.
Authors: Lukas Helff, Felix Friedrich, Manuel Brack, Kristian Kersting, Patrick Schramowski
Abstract: This paper introduces LlavaGuard, a suite of VLM-based vision safeguards that address the critical need for reliable guardrails in the era of large-scale data and models. To this end, we establish a novel open framework, describing a customizable safety taxonomy, data preprocessing, augmentation, and training setup. For teaching a VLM safeguard on safety, we further create a multimodal safety dataset with high-quality human expert annotations, where each image is labeled with a safety rating, category and rationale. We also employ advanced augmentations to support context-specific assessments. The resulting LlavaGuard models, ranging from 0.5B to 7B, serve as a versatile tool for evaluating the safety compliance of visual content against flexible policies. In comprehensive experiments, LlavaGuard outperforms both state-of-the-art safeguards and VLMs in accuracy and in flexibly handling different policies. Additionally, we demonstrate LlavaGuard's performance in two real-world applications: large-scale dataset annotation and moderation of text-to-image models. We make our entire framework publicly available, including the dataset and model weights.
Authors: Theshani Nuradha, Mark M. Wilde
Abstract: A quantum generalized divergence by definition satisfies the data-processing inequality; as such, the relative decrease in such a divergence under the action of a quantum channel is at most one. This relative decrease is formally known as the contraction coefficient of the channel and the divergence. Interestingly, there exist combinations of channels and divergences for which the contraction coefficient is strictly less than one. Furthermore, understanding the contraction coefficient is fundamental for the study of statistical tasks under privacy constraints. To this end, here we establish upper bounds on contraction coefficients for the hockey-stick divergence under privacy constraints, where privacy is quantified with respect to the quantum local differential privacy (QLDP) framework, and we fully characterize the contraction coefficient for the trace distance under privacy constraints. With the machinery developed, we also determine an upper bound on the contraction of both the Bures distance and quantum relative entropy relative to the normalized trace distance, under QLDP constraints. Next, we apply our findings to establish bounds on the sample complexity of quantum hypothesis testing under privacy constraints. Furthermore, we study various scenarios in which the sample complexity bounds are tight, while providing order-optimal quantum channels that achieve those bounds. Lastly, we show how private quantum channels provide fairness and Holevo information stability in quantum learning settings.
Authors: Iveta Be\v{c}kov\'a, \v{S}tefan P\'oco\v{s}, Giulia Belgiovine, Marco Matarese, Omar Eldardeer, Alessandra Sciutti, Carlo Mazzola
Abstract: The addressee estimation (understanding to whom somebody is talking) is a fundamental task for human activity recognition in multi-party conversation scenarios. Specifically, in the field of human-robot interaction, it becomes even more crucial to enable social robots to participate in such interactive contexts. However, it is usually implemented as a binary classification task, restricting the robot's capability to estimate whether it was addressed \review{or not, which} limits its interactive skills. For a social robot to gain the trust of humans, it is also important to manifest a certain level of transparency and explainability. Explainable artificial intelligence thus plays a significant role in the current machine learning applications and models, to provide explanations for their decisions besides excellent performance. In our work, we a) present an addressee estimation model with improved performance in comparison with the previous state-of-the-art; b) further modify this model to include inherently explainable attention-based segments; c) implement the explainable addressee estimation as part of a modular cognitive architecture for multi-party conversation in an iCub robot; d) validate the real-time performance of the explainable model in multi-party human-robot interaction; e) propose several ways to incorporate explainability and transparency in the aforementioned architecture; and f) perform an online user study to analyze the effect of various explanations on how human participants perceive the robot.
Authors: Guokai Li, Zizhuo Wang, Jingwei Zhang
Abstract: Online linear programming (OLP) has gained significant attention from both researchers and practitioners due to its extensive applications, such as online auction, network revenue management, order fulfillment and advertising. Existing OLP algorithms fall into two categories: LP-based algorithms and LP-free algorithms. The former one typically guarantees better performance, even offering a constant regret, but requires solving a large number of LPs, which could be computationally expensive. In contrast, LP-free algorithm only requires first-order computations but induces a worse performance, lacking a constant regret bound. In this work, we bridge the gap between these two extremes by proposing a well-performing algorithm, that solves LPs at a few selected time points and conducts first-order computations at other time points. Specifically, for the case where the inputs are drawn from an unknown finite-support distribution, the proposed algorithm achieves a constant regret (even for the hard "degenerate" case) while solving LPs only $\mathcal{O}(\log\log T)$ times over the time horizon $T$. Moreover, when we are allowed to solve LPs only $M$ times, we design the corresponding schedule such that the proposed algorithm can guarantee a nearly $\mathcal{O}\left(T^{(1/2)^{M-1}}\right)$ regret. Our work highlights the value of resolving both at the beginning and the end of the selling horizon, and provides a novel framework to prove the performance guarantee of the proposed policy under different infrequent resolving schedules. Furthermore, when the arrival probabilities are known at the beginning, our algorithm can guarantee a constant regret by solving LPs $\mathcal{O}(\log\log T)$ times, and a nearly $\mathcal{O}\left(T^{(1/2)^{M}}\right)$ regret by solving LPs only $M$ times. Numerical experiments are conducted to demonstrate the efficiency of the proposed algorithms.
Authors: Gaole Dai, Chun-Kai Fan, Yiming Tang, Zhi Zhang, Yuan Zhang, Yulu Gan, Qizhe Zhang, Cheng-Ching Tseng, Shanghang Zhang, Tiejun Huang
Abstract: Advances in Parameter-Efficient Fine-Tuning (PEFT) bridged the performance gap with Full Fine-Tuning (FFT) through sophisticated analysis of pre-trained parameter spaces. Starting from drawing insights from Neural Engrams (NE) in Biological Neural Networks (BNNs), we establish a connection between the low-rank property observed during PEFT's parameter space shifting and neurobiological mechanisms. This observation leads to our proposed method, Synapse and Neuron (SAN), which decomposes and propagates scaling components from anterior feature adjusting vectors towards posterior weight matrices. Our approach is theoretically grounded in Long-Term Potentiation/Depression (LTP/D) phenomena, which govern synapse development through neurotransmitter release modulation. Extensive experiments demonstrate its effectiveness: on \textbf{vision tasks} across VTAB, FGVC, and GIC (25 datasets) using ViT, SwinT and ConvNeXt, SAN outperforms FFT up to 8.7% and LoRA by 3.2%; on language tasks using Commonsense Reasoning (8 datasets) with LLaMA models (all generations), surpassing ChatGPT up to 8.5% and LoRA by 4.7%; on visual-language tasks using Mixed Visual Instruction (7 datasets) with LLaVA models, it exceeds FFT up to 2.4% and LoRA by 1.9%. Our code and W&B log will be released in https://github.com/daviddaiiiii/SAN-PEFT
Authors: Sam Gijsen, Kerstin Ritter
Abstract: Multimodal language modeling has enabled breakthroughs for representation learning, yet remains unexplored in the realm of functional brain data for pathology detection. This paper pioneers EEG-language models (ELMs) trained on clinical reports and 15000 EEGs. We propose to combine multimodal alignment in this novel domain with timeseries cropping and text segmentation, enabling an extension based on multiple instance learning to alleviate misalignment between irrelevant EEG or text segments. Our multimodal models significantly improve pathology detection compared to EEG-only models across four evaluations and for the first time enable zero-shot classification as well as retrieval of both neural signals and reports. In sum, these results highlight the potential of ELMs, representing significant progress for clinical applications.
Authors: Daoyang Li, Haiyan Zhao, Qingcheng Zeng, Mengnan Du
Abstract: Probing techniques for large language models (LLMs) have primarily focused on English, overlooking the vast majority of the world's languages. In this paper, we extend these probing methods to a multilingual context, investigating the behaviors of LLMs across diverse languages. We conduct experiments on several open-source LLM models, analyzing probing accuracy, trends across layers, and similarities between probing vectors for multiple languages. Our key findings reveal: (1) a consistent performance gap between high-resource and low-resource languages, with high-resource languages achieving significantly higher probing accuracy; (2) divergent layer-wise accuracy trends, where high-resource languages show substantial improvement in deeper layers similar to English; and (3) higher representational similarities among high-resource languages, with low-resource languages demonstrating lower similarities both among themselves and with high-resource languages. These results highlight significant disparities in LLMs' multilingual capabilities and emphasize the need for improved modeling of low-resource languages.
Authors: Matteo Zecchin, Sangwoo Park, Osvaldo Simeone
Abstract: We introduce adaptive learn-then-test (aLTT), an efficient hyperparameter selection procedure that provides finite-sample statistical guarantees on the population risk of AI models. Unlike the existing learn-then-test (LTT) technique, which relies on conventional p-value-based multiple hypothesis testing (MHT), aLTT implements sequential data-dependent MHT with early termination by leveraging e-processes. As a result, aLTT can reduce the number of testing rounds, making it particularly well-suited for scenarios in which testing is costly or presents safety risks. Apart from maintaining statistical validity, in applications such as online policy selection for offline reinforcement learning and prompt engineering, aLTT is shown to achieve the same performance as LTT while requiring only a fraction of the testing rounds.
Authors: Adrian Pekar
Abstract: Network monitoring generates massive volumes of IP flow records, posing significant challenges for storage and analysis. This paper presents a novel deep learning-based approach to compressing these records using autoencoders, enabling direct analysis of compressed data without requiring decompression. Unlike traditional compression methods, our approach reduces data volume while retaining the utility of compressed data for downstream analysis tasks, including distinguishing modern application protocols and encrypted traffic from popular services. Through extensive experiments on a real-world network traffic dataset, we demonstrate that our autoencoder-based compression achieves a 1.313x reduction in data size while maintaining 99.27% accuracy in a multi-class traffic classification task, compared to 99.77% accuracy with uncompressed data. This marginal decrease in performance is offset by substantial gains in storage and processing efficiency. The implications of this work extend to more efficient network monitoring and scalable, real-time network management solutions.
Authors: Jifan Zhang, Ziyue Luo, Jia Liu, Ness Shroff, Robert Nowak
Abstract: Large language models require vast amounts of high-quality training data, but effective filtering of web-scale datasets remains a significant challenge. This paper demonstrates that GPT-4o is remarkably effective at identifying high-quality training data, but its prohibitive cost makes it impractical at web-scale. We propose SIEVE, a lightweight alternative that matches GPT-4o accuracy at less than 1\% of the cost. SIEVE can perform up to 500 filtering operations for the cost of one GPT-4o filtering call. The key to SIEVE is a seamless integration of GPT-4o and lightweight text classification models, using active learning to fine-tune these models in the background with a small number of calls to GPT-4o. Once trained, it performs as well as GPT-4o at a tiny fraction of the cost. Through different filtering prompts, SIEVE can efficiently curate high quality data for general or specialized domains from web-scale corpora -- a valuable capability given the current scarcity of high-quality domain-specific datasets. Extensive experiments using automatic and human evaluation metrics show that SIEVE and GPT-4o achieve similar performance on five highly specific filtering prompts. In addition, when performing quality filtering on web crawl datasets, we demonstrate SIEVE can further improve over state-of-the-art quality filtering methods in the DataComp-LM challenge for selecting LLM pretraining data.
Authors: Judah Goldfeder, Victoria Dean, Zixin Jiang, Xuezheng Wang, Bing dong, Hod Lipson, John Sipple
Abstract: Commercial buildings account for 17% of U.S. carbon emissions, with roughly half of that from Heating, Ventilation, and Air Conditioning (HVAC). HVAC devices form a complex thermodynamic system, and while Model Predictive Control and Reinforcement Learning have been used to optimize control policies, scaling to thousands of buildings remains a significant unsolved challenge. Most current algorithms are over-optimized for specific buildings and rely on proprietary data or hard-to-configure simulations. We present the Smart Buildings Control Suite, the first open source interactive HVAC control benchmark with a focus on solutions that scale. It consists of 3 components: real-world telemetric data extracted from 11 buildings over 6 years, a lightweight data-driven simulator for each building, and a modular Physically Informed Neural Network (PINN) building model as a simulator alternative. The buildings span a variety of climates, management systems, and sizes, and both the simulator and PINN easily scale to new buildings, ensuring solutions using this benchmark are robust to these factors and only reliant on fully scalable building models. This represents a major step towards scaling HVAC optimization from the lab to buildings everywhere. To facilitate use, our benchmark is compatible with the Gym standard, and our data is part of TensorFlow Datasets.
Authors: Giovanni Monea, Antoine Bosselut, Kiant\'e Brantley, Yoav Artzi
Abstract: Large Language Models (LLMs) excel at in-context learning (ICL), a supervised learning technique that relies on adding annotated examples to the model context. We investigate a contextual bandit version of in-context reinforcement learning (ICRL), where models learn in-context, online, from external reward, instead of supervised data. We show that LLMs effectively demonstrate such learning, and provide a detailed study of the phenomena, experimenting with challenging classification tasks and models of sizes from 500M to 70B parameters. This includes identifying and addressing the instability of the process, demonstrating learning with both semantic and abstract labels, and showing scaling trends. Our findings highlight ICRL capabilities in LLMs, while also underscoring fundamental limitations in their implicit reasoning about errors.
Authors: Vincent Zhihao Zheng, Lijun Sun
Abstract: Multivariate Gaussian (MVG) distributions are central to modeling correlated continuous variables in probabilistic forecasting. Neural forecasting models typically parameterize the mean vector and covariance matrix of the distribution using neural networks, optimizing with the log-score (negative log-likelihood) as the loss function. However, the sensitivity of the log-score to outliers can lead to significant errors in the presence of anomalies. Drawing on the continuous ranked probability score (CRPS) for univariate distributions, we propose MVG-CRPS, a strictly proper scoring rule for MVG distributions. MVG-CRPS admits a closed-form expression in terms of neural network outputs, thereby integrating seamlessly into deep learning frameworks. Experiments on real-world datasets across multivariate autoregressive and univariate sequence-to-sequence (Seq2Seq) forecasting tasks show that MVG-CRPS improves robustness, accuracy, and uncertainty quantification in probabilistic forecasting.
Authors: Zhengrui Guo, Fangxu Zhou, Wei Wu, Qichen Sun, Lishuang Feng, Jinzhuo Wang, Hao Chen
Abstract: Modeling the nonlinear dynamics of neuronal populations represents a key pursuit in computational neuroscience. Recent research has increasingly focused on jointly modeling neural activity and behavior to unravel their interconnections. Despite significant efforts, these approaches often necessitate either intricate model designs or oversimplified assumptions. Given the frequent absence of perfectly paired neural-behavioral datasets in real-world scenarios when deploying these models, a critical yet understudied research question emerges: how to develop a model that performs well using only neural activity as input at inference, while benefiting from the insights gained from behavioral signals during training? To this end, we propose BLEND, the behavior-guided neural population dynamics modeling framework via privileged knowledge distillation. By considering behavior as privileged information, we train a teacher model that takes both behavior observations (privileged features) and neural activities (regular features) as inputs. A student model is then distilled using only neural activity. Unlike existing methods, our framework is model-agnostic and avoids making strong assumptions about the relationship between behavior and neural activity. This allows BLEND to enhance existing neural dynamics modeling architectures without developing specialized models from scratch. Extensive experiments across neural population activity modeling and transcriptomic neuron identity prediction tasks demonstrate strong capabilities of BLEND, reporting over 50% improvement in behavioral decoding and over 15% improvement in transcriptomic neuron identity prediction after behavior-guided distillation. Furthermore, we empirically explore various behavior-guided distillation strategies within the BLEND framework and present a comprehensive analysis of effectiveness and implications for model performance.
Authors: Eric Elmoznino, Thomas Jiralerspong, Yoshua Bengio, Guillaume Lajoie
Abstract: Compositionality is believed to be fundamental to intelligence. In humans, it underlies the structure of thought, language, and higher-level reasoning. In AI, compositional representations can enable a powerful form of out-of-distribution generalization, in which a model systematically adapts to novel combinations of known concepts. However, while we have strong intuitions about what compositionality is, there currently exists no formal definition for it that is measurable and mathematical. Here, we propose such a definition, which we call representational compositionality, that accounts for and extends our intuitions about compositionality. The definition is conceptually simple, quantitative, grounded in algorithmic information theory, and applicable to any representation. Intuitively, representational compositionality states that a compositional representation satisfies three properties. First, it must be expressive. Second, it must be possible to re-describe the representation as a function of discrete symbolic sequences with re-combinable parts, analogous to sentences in natural language. Third, the function that relates these symbolic sequences to the representation, analogous to semantics in natural language, must be simple. Through experiments on both synthetic and real world data, we validate our definition of compositionality and show how it unifies disparate intuitions from across the literature in both AI and cognitive science. We also show that representational compositionality, while theoretically intractable, can be readily estimated using standard deep learning tools. Our definition has the potential to inspire the design of novel, theoretically-driven models that better capture the mechanisms of compositional thought.
Authors: Leni Aniva, Chuyue Sun, Brando Miranda, Clark Barrett, Sanmi Koyejo
Abstract: Machine-assisted theorem proving refers to the process of conducting structured reasoning to automatically generate proofs for mathematical theorems. Recently, there has been a surge of interest in using machine learning models in conjunction with proof assistants to perform this task. In this paper, we introduce Pantograph, a tool that provides a versatile interface to the Lean 4 proof assistant and enables efficient proof search via powerful search algorithms such as Monte Carlo Tree Search. In addition, Pantograph enables high-level reasoning by enabling a more robust handling of Lean 4's inference steps. We provide an overview of Pantograph's architecture and features. We also report on an illustrative use case: using machine learning models and proof sketches to prove Lean 4 theorems. Pantograph's innovative features pave the way for more advanced machine learning models to perform complex proof searches and high-level reasoning, equipping future researchers to design more versatile and powerful theorem provers.
Authors: Chao Yu, Qixin Tan, Hong Lu, Jiaxuan Gao, Xinting Yang, Yu Wang, Yi Wu, Eugene Vinitsky
Abstract: Preference-based reinforcement learning is an effective way to handle tasks where rewards are hard to specify but can be exceedingly inefficient as preference learning is often tabula rasa. We demonstrate that Large Language Models (LLMs) have native preference-learning capabilities that allow them to achieve sample-efficient preference learning, addressing this challenge. We propose In-Context Preference Learning (ICPL), which uses in-context learning capabilities of LLMs to reduce human query inefficiency. ICPL uses the task description and basic environment code to create sets of reward functions which are iteratively refined by placing human feedback over videos of the resultant policies into the context of an LLM and then requesting better rewards. We first demonstrate ICPL's effectiveness through a synthetic preference study, providing quantitative evidence that it significantly outperforms baseline preference-based methods with much higher performance and orders of magnitude greater efficiency. We observe that these improvements are not solely coming from LLM grounding in the task but that the quality of the rewards improves over time, indicating preference learning capabilities. Additionally, we perform a series of real human preference-learning trials and observe that ICPL extends beyond synthetic settings and can work effectively with humans-in-the-loop.
Authors: Sam Blouir, Jimmy T. H. Smith, Antonios Anastasopoulos, Amarda Shehu
Abstract: Efficient state space models (SSMs), such as linear recurrent neural networks and linear attention variants, offer computational advantages over Transformers but struggle with tasks requiring long-range in-context retrieval-like text copying, associative recall, and question answering over long contexts. Previous efforts to address these challenges have focused on architectural modifications, often reintroducing computational inefficiencies. In this paper, we propose a novel training procedure, Birdie, that significantly enhances the in-context retrieval capabilities of SSMs without altering their architecture. Our approach combines bidirectional input processing with dynamic mixtures of specialized pre-training objectives, optimized via reinforcement learning. We introduce a new bidirectional SSM architecture that seamlessly transitions from bidirectional context processing to causal generation. Experimental evaluations demonstrate that Birdie markedly improves performance on retrieval-intensive tasks such as multi-number phone book lookup, long paragraph question-answering, and infilling. This narrows the performance gap with Transformers, while retaining computational efficiency. Our findings highlight the importance of training procedures in leveraging the fixed-state capacity of SSMs, offering a new direction to advance their capabilities. All code and pre-trained models are available at https://www.github.com/samblouir/birdie, with support for JAX and PyTorch.
Authors: Sejoon Oh, Yiqiao Jin, Megha Sharma, Donghyun Kim, Eric Ma, Gaurav Verma, Srijan Kumar
Abstract: Multimodal large language models (MLLMs) have revolutionized vision-language understanding but remain vulnerable to multimodal jailbreak attacks, where adversarial inputs are meticulously crafted to elicit harmful or inappropriate responses. We propose UniGuard, a novel multimodal safety guardrail that jointly considers the unimodal and cross-modal harmful signals. UniGuard trains a multimodal guardrail to minimize the likelihood of generating harmful responses in a toxic corpus. The guardrail can be seamlessly applied to any input prompt during inference with minimal computational costs. Extensive experiments demonstrate the generalizability of UniGuard across multiple modalities, attack strategies, and multiple state-of-the-art MLLMs, including LLaVA, Gemini Pro, GPT-4o, MiniGPT-4, and InstructBLIP. Notably, this robust defense mechanism maintains the models' overall vision-language understanding capabilities.
Authors: David Chapman, Parniyan Farvardin
Abstract: We present a novel empirical approach toward measuring the Probability Density Function (PDF) of the deep features of Convolutional Neural Networks (CNNs). Measurement of the deep feature PDF is a valuable problem for several reasons. Notably, a. Understanding the deep feature PDF yields new insight into deep representations. b. Feature density methods are important for tasks such as anomaly detection which can improve the robustness of deep learning models in the wild. Interpretable measurement of the deep feature PDF is challenging due to the Curse of Dimensionality (CoD), and the Spatial intuition Limitation. Our novel measurement technique combines copula analysis with the Method of Orthogonal Moments (MOM), in order to directly measure the Generalized Characteristic Function (GCF) of the multivariate deep feature PDF. We find that, surprisingly, the one-dimensional marginals of non-negative deep CNN features after major blocks are not well approximated by a Gaussian distribution, and that these features increasingly approximate an exponential distribution with increasing network depth. Furthermore, we observe that deep features become increasingly independent with increasing network depth within their typical ranges. However, we surprisingly also observe that many deep features exhibit strong dependence (either correlation or anti-correlation) with other extremely strong detections, even if these features are independent within typical ranges. We elaborate on these findings in our discussion, where we propose a new hypothesis that exponentially infrequent large valued features correspond to strong computer vision detections of semantic targets, which would imply that these large-valued features are not outliers but rather an important detection signal.
Authors: Ali Marashian, Enora Rice, Luke Gessler, Alexis Palmer, Katharina von der Wense
Abstract: Many of the world's languages have insufficient data to train high-performing general neural machine translation (NMT) models, let alone domain-specific models, and often the only available parallel data are small amounts of religious texts. Hence, domain adaptation (DA) is a crucial issue faced by contemporary NMT and has, so far, been underexplored for low-resource languages. In this paper, we evaluate a set of methods from both low-resource NMT and DA in a realistic setting, in which we aim to translate between a high-resource and a low-resource language with access to only: a) parallel Bible data, b) a bilingual dictionary, and c) a monolingual target-domain corpus in the high-resource language. Our results show that the effectiveness of the tested methods varies, with the simplest one, DALI, being most effective. We follow up with a small human evaluation of DALI, which shows that there is still a need for more careful investigation of how to accomplish DA for low-resource NMT.
Authors: Ishan Amin, Sanjeev Raja, Aditi Krishnapriyan
Abstract: The foundation model (FM) paradigm is transforming Machine Learning Force Fields (MLFFs), leveraging general-purpose representations and scalable training to perform a variety of computational chemistry tasks. Although MLFF FMs have begun to close the accuracy gap relative to first-principles methods, there is still a strong need for faster inference speed. Additionally, while research is increasingly focused on general-purpose models which transfer across chemical space, practitioners typically only study a small subset of systems at a given time. This underscores the need for fast, specialized MLFFs relevant to specific downstream applications, which preserve test-time physical soundness while maintaining train-time scalability. In this work, we introduce a method for transferring general-purpose representations from MLFF foundation models to smaller, faster MLFFs specialized to specific regions of chemical space. We formulate our approach as a knowledge distillation procedure, where the smaller "student" MLFF is trained to match the Hessians of the energy predictions of the "teacher" foundation model. Our specialized MLFFs can be up to 20 $\times$ faster than the original foundation model, while retaining, and in some cases exceeding, its performance and that of undistilled models. We also show that distilling from a teacher model with a direct force parameterization into a student model trained with conservative forces (i.e., computed as derivatives of the potential energy) successfully leverages the representations from the large-scale teacher for improved accuracy, while maintaining energy conservation during test-time molecular dynamics simulations. More broadly, our work suggests a new paradigm for MLFF development, in which foundation models are released along with smaller, specialized simulation "engines" for common chemical subsets.
Authors: Kangjie Zheng, Junwei Yang, Siyue Liang, Bin Feng, Zequn Liu, Wei Ju, Zhiping Xiao, Ming Zhang
Abstract: Masked Language Models (MLMs) have achieved remarkable success in many self-supervised representation learning tasks. MLMs are trained by randomly masking portions of the input sequences with [MASK] tokens and learning to reconstruct the original content based on the remaining context. This paper explores the impact of [MASK] tokens on MLMs. Analytical studies show that masking tokens can introduce the corrupted semantics problem, wherein the corrupted context may convey multiple, ambiguous meanings. This problem is also a key factor affecting the performance of MLMs on downstream tasks. Based on these findings, we propose a novel enhanced-context MLM, ExLM. Our approach expands [MASK] tokens in the input context and models the dependencies between these expanded states. This enhancement increases context capacity and enables the model to capture richer semantic information, effectively mitigating the corrupted semantics problem during pre-training. Experimental results demonstrate that ExLM achieves significant performance improvements in both text modeling and SMILES modeling tasks. Further analysis confirms that ExLM enriches semantic representations through context enhancement, and effectively reduces the semantic multimodality commonly observed in MLMs.
Authors: Oliver Goldstein, Samuel March
Abstract: The Algebraic Data Type (ADT) can be used as a computational framework for molecular representation for the purpose of advancing tasks in cheminformatics. This can include generative modles in the context of Bayesian machine learning via probabilistic programming. The ADT that we put forward, implements the 'Dietz' representation for molecular constitution via multigraphs of electron valence information, and uses 3D coordinate data to provide stereochemical information, easily enabling the representation of complex molecular phenomena such as organometallics, multi-center bonds, delocalized electrons, and resonant structures. Unlike traditional string-based representations such as SMILES and SELFIES, the ADT is much more flexible, yet retains desirable qualities from type-safety to seamless integration with Bayesian Probabilistic Programming. An extensive criticism of both SMILES and SELFIES, in this article, is given, along with criticisms of the so-called Future of SELFIES. An open-source library implemented in Haskell demonstrates the ADT along with experimental extensions demonstrating its use in reaction modelling, group theoretic applications, and integration with LazyPPL, a lazy probabilistic programming library. Also provided as an extension is the ability to represent electronic structures, including shells, subshells, and orbitals, significantly expanding its representational scope compared to other representations in the literature. These features position the proposed ADT as a robust alternative to existing molecular representations, addressing limitations such as inadequate support for 3D information and syntactic invalidity while offering a platform for innovative cheminformatics research. Accompanying discussions about the meaning of a `representation' are included. The fully functioning GitHub library can be found at https://github.com/oliverjgoldstein/chemalgprog.
Authors: James Seale Smith, Chi-Heng Lin, Shikhar Tuli, Haris Jeelani, Shangqian Gao, Yilin Shen, Hongxia Jin, Yen-Chang Hsu
Abstract: The rapid proliferation of large language models (LLMs) in natural language processing (NLP) has created a critical need for techniques that enable efficient deployment on memory-constrained devices without compromising performance. We present a method to prune LLMs that selectively prunes model blocks based on an importance score and replaces them with a low-parameter replacement strategy. Specifically, we propose a principled metric to replace each pruned block using a weight-sharing mechanism that leverages unpruned counterparts from the model and block-specific low-rank adapters. Furthermore, we facilitate the learning of these replacement blocks with output feature normalization and an adapter initialization scheme built on low-rank SVD reconstructions. Empirical evaluations demonstrate substantial performance gains over existing methods, achieving state-of-the-art performance on 5/6 benchmarks for a compression rate of 30% and 6/6 benchmarks for a compression rate of 40%. We also demonstrate that our approach can extend smaller models, boosting performance on 6/6 benchmarks using only ~0.3% tokens of extended training with minimal additional parameter costs.
Authors: Francesco Della Santa, Morgana Lalli
Abstract: This study presents a novel Deep Learning-based and lightweight approach for the automated detection of sports highlights (HLs) from audio and video sources. HL detection is a key task in sports video analysis, traditionally requiring significant human effort. Our solution leverages Deep Learning (DL) models trained on relatively small datasets of audio Mel-spectrograms and grayscale video frames, achieving promising accuracy rates of 89% and 83% for audio and video detection, respectively. The use of small datasets, combined with simple architectures, demonstrates the practicality of our method for fast and cost-effective deployment. Furthermore, an ensemble model combining both modalities shows improved robustness against false positives and false negatives. The proposed methodology offers a scalable solution for automated HL detection across various types of sports video content, reducing the need for manual intervention. Future work will focus on enhancing model architectures and extending this approach to broader scene-detection tasks in media analysis.
Authors: Mateusz Nowak, Wojciech Jarosz, Peter Chin
Abstract: Reconstructing a 3D scene from images is challenging due to the different ways light interacts with surfaces depending on the viewer's position and the surface's material. In classical computer graphics, materials can be classified as diffuse or specular, interacting with light differently. The standard 3D Gaussian Splatting model struggles to represent view-dependent content, since it cannot differentiate an object within the scene from the light interacting with its specular surfaces, which produce highlights or reflections. In this paper, we propose to extend the 3D Gaussian Splatting model by introducing an additional symmetric matrix to enhance the opacity representation of each 3D Gaussian. This improvement allows certain Gaussians to be suppressed based on the viewer's perspective, resulting in a more accurate representation of view-dependent reflections and specular highlights without compromising the scene's integrity. By allowing the opacity to be view dependent, our enhanced model achieves state-of-the-art performance on Mip-Nerf, Tanks&Temples, Deep Blending, and Nerf-Synthetic datasets without a significant loss in rendering speed, achieving >60FPS, and only incurring a minimal increase in memory used.