Authors: Ganesh Parab, Zishan Ahmad, Dagnachew Birru
Abstract: Text-to-SQL generation bridges the gap between natural language and databases, enabling users to query data without requiring SQL expertise. While large language models (LLMs) have significantly advanced the field, challenges remain in handling complex queries that involve multi-table joins, nested conditions, and intricate operations. Existing methods often rely on multi-step pipelines that incur high computational costs, increase latency, and are prone to error propagation. To address these limitations, we propose HI-SQL, a pipeline that incorporates a novel hint generation mechanism utilizing historical query logs to guide SQL generation. By analyzing prior queries, our method generates contextual hints that focus on handling the complexities of multi-table and nested operations. These hints are seamlessly integrated into the SQL generation process, eliminating the need for costly multi-step approaches and reducing reliance on human-crafted prompts. Experimental evaluations on multiple benchmark datasets demonstrate that our approach significantly improves query accuracy of LLM-generated queries while ensuring efficiency in terms of LLM calls and latency, offering a robust and practical solution for enhancing Text-to-SQL systems.
Authors: Shriyank Somvanshi, Md Monzurul Islam, Gaurab Chhetri, Rohit Chakraborty, Mahmuda Sultana Mimi, Swagat Ahmed Shuvo, Kazi Sifatul Islam, Syed Aaqib Javed, Sharif Ahmed Rafat, Anandi Dutta, Subasish Das
Abstract: The rapid growth of edge devices has driven the demand for deploying artificial intelligence (AI) at the edge, giving rise to Tiny Machine Learning (TinyML) and its evolving counterpart, Tiny Deep Learning (TinyDL). While TinyML initially focused on enabling simple inference tasks on microcontrollers, the emergence of TinyDL marks a paradigm shift toward deploying deep learning models on severely resource-constrained hardware. This survey presents a comprehensive overview of the transition from TinyML to TinyDL, encompassing architectural innovations, hardware platforms, model optimization techniques, and software toolchains. We analyze state-of-the-art methods in quantization, pruning, and neural architecture search (NAS), and examine hardware trends from MCUs to dedicated neural accelerators. Furthermore, we categorize software deployment frameworks, compilers, and AutoML tools enabling practical on-device learning. Applications across domains such as computer vision, audio recognition, healthcare, and industrial monitoring are reviewed to illustrate the real-world impact of TinyDL. Finally, we identify emerging directions including neuromorphic computing, federated TinyDL, edge-native foundation models, and domain-specific co-design approaches. This survey aims to serve as a foundational resource for researchers and practitioners, offering a holistic view of the ecosystem and laying the groundwork for future advancements in edge AI.
Authors: Shuang Ao, Yi Dong, Jinwei Hu, Sarvapali Ramchurn
Abstract: Fine-tuning Large Language Models (LLMs) with Low-Rank Adaptation (LoRA) enhances adaptability while reducing computational costs. However, fine-tuning can compromise safety alignment, even with benign data, increasing susceptibility to harmful outputs. Existing safety alignment methods struggle to capture complex parameter shifts, leading to suboptimal safety-utility trade-offs. To address this issue, we propose Safe Pruning LoRA (SPLoRA), a novel pruning-based approach that selectively removes LoRA layers that weaken safety alignment, improving safety while preserving performance. At its core, we introduce Empirical-DIEM (E-DIEM), a dimension-insensitive similarity metric that effectively detects safety misalignment in LoRA-adapted models. We conduct extensive experiments on LLMs fine-tuned with mixed of benign and malicious data, and purely benign datasets, evaluating SPLoRA across utility, safety, and reliability metrics. Results demonstrate that SPLoRA outperforms state-of-the-art safety alignment techniques, significantly reducing safety risks while maintaining or improving model performance and reliability. Additionally, SPLoRA reduces inference overhead, making it a scalable and efficient solution for deploying safer and more reliable LLMs. The code is available at https://github.com/AoShuang92/SPLoRA.
Authors: Zihan Wang, Rui Pan, Jiarui Yao, Robert Csordas, Linjie Li, Lu Yin, Jiajun Wu, Tong Zhang, Manling Li, Shiwei Liu
Abstract: We propose Chain-of-Experts (CoE), a new Mixture-of-Experts (MoE) architecture that introduces sequential expert communication within each layer. Unlike traditional MoE models, where experts operate independently in parallel, CoE processes tokens iteratively across a chain of experts inside a layer. To support dynamic expert selection across iterations, CoE employs a dedicated router at each iteration step within a layer. This design allows tokens to re-evaluate and select different experts during each iteration, rather than being statically assigned. As a result, CoE introduces a flexible routing mechanism that increases the diversity of expert combinations and enriches the model's representational capacity. CoE demonstrates improved performance under fixed compute: on math reasoning tasks, it reduces validation loss from 1.20 to 1.12 compared to a standard MoE. Beyond performance, CoE offers a new scaling axis: depth through expert iteration, which complements conventional width/depth scaling. For example, using 2x iterations matches the performance of 3x expert selections (in width), while reducing memory usage by 17.6-42% relative to other scaling strategies. Our analysis reveals that CoE's benefits stem from its iterative residual structure and enhanced expert specialization empowered by iterative routing, which together unlock more expressive representations. Code is available at https://github.com/ZihanWang314/coe.
Authors: Maoyuan Li, Sihong Li, Guancheng Shen, Yun Zhang, Huamin Zhou
Abstract: To address the challenges of untimely detection and online monitoring lag in injection molding quality anomalies, this study proposes a mixed feature attention-artificial neural network (MFA-ANN) model for high-precision online prediction of product weight. By integrating mechanism-based with data-driven analysis, the proposed architecture decouples time series data (e.g., melt flow dynamics, thermal profiles) from non-time series data (e.g., mold features, pressure settings), enabling hierarchical feature extraction. A self-attention mechanism is strategically embedded during cross-domain feature fusion to dynamically calibrate inter-modality feature weights, thereby emphasizing critical determinants of weight variability. The results demonstrate that the MFA-ANN model achieves a RMSE of 0.0281 with 0.5 g weight fluctuation tolerance, outperforming conventional benchmarks: a 25.1% accuracy improvement over non-time series ANN models, 23.0% over LSTM networks, 25.7% over SVR, and 15.6% over RF models, respectively. Ablation studies quantitatively validate the synergistic enhancement derived from the integration of mixed feature modeling (contributing 22.4%) and the attention mechanism (contributing 11.2%), significantly enhancing the model's adaptability to varying working conditions and its resistance to noise. Moreover, critical sensitivity analyses further reveal that data resolution significantly impacts prediction reliability, low-fidelity sensor inputs degrade performance by 23.8% RMSE compared to high-precision measurements. Overall, this study provides an efficient and reliable solution for the intelligent quality control of injection molding processes.
Authors: Zohaib Hasan Siddiqui, Jiechao Gao, Ebad Shabbir, Mohammad Anas Azeez, Rafiq Ali, Gautam Siddharth Kashyap, Usman Naseem
Abstract: Running Large Language Models (LLMs) on edge devices is constrained by high compute and memory demands posing a barrier for real-time applications in sectors like healthcare, education, and embedded systems. Current solutions such as quantization, pruning, and retrieval-augmented generation (RAG) offer only partial optimizations and often compromise on speed or accuracy. We introduce HOLA, an end-to-end optimization framework for efficient LLM deployment. Internally, it leverages Hierarchical Speculative Decoding (HSD) for faster inference without quality loss. Externally, AdaComp-RAG adjusts retrieval complexity based on context needs. Together with LoBi, which blends structured pruning (LoRA) and quantization, HOLA delivers significant gains: 17.6% EMA on GSM8K, 10.5% MCA on ARC, and reduced latency and memory on edge devices like Jetson Nano--proving both scalable and production-ready.
Authors: Hanshuo Wu, Xudong Jian, Christos Lataniotis, Cyprien Hoelzl, Eleni Chatzi, Yves Reuland
Abstract: Bridges, as critical components of civil infrastructure, are increasingly affected by deterioration, making reliable traffic monitoring essential for assessing their remaining service life. Among operational loads, traffic load plays a pivotal role, and recent advances in deep learning - particularly in computer vision (CV) - have enabled progress toward continuous, automated monitoring. However, CV-based approaches suffer from limitations, including privacy concerns and sensitivity to lighting conditions, while traditional non-vision-based methods often lack flexibility in deployment and validation. To bridge this gap, we propose a fully automated deep-learning pipeline for continuous traffic monitoring using structural health monitoring (SHM) sensor networks. Our approach integrates CV-assisted high-resolution dataset generation with supervised training and inference, leveraging graph neural networks (GNNs) to capture the spatial structure and interdependence of sensor data. By transferring knowledge from CV outputs to SHM sensors, the proposed framework enables sensor networks to achieve comparable accuracy of vision-based systems, with minimal human intervention. Applied to accelerometer and strain gauge data in a real-world case study, the model achieves state-of-the-art performance, with classification accuracies of 99% for light vehicles and 94% for heavy vehicles.
Authors: Shashank Yadav, Vignesh Subbian
Abstract: Interpretability plays a vital role in aligning and deploying deep learning models in critical care, especially in constantly evolving conditions that influence patient survival. However, common interpretability algorithms face unique challenges when applied to dynamic prediction tasks, where patient trajectories evolve over time. Gradient, Occlusion, and Permutation-based methods often struggle with time-varying target dependency and temporal smoothness. This work systematically analyzes these failure modes and supports learnable mask-based interpretability frameworks as alternatives, which can incorporate temporal continuity and label consistency constraints to learn feature importance over time. Here, we propose that learnable mask-based approaches for dynamic timeseries prediction problems provide more reliable and consistent interpretations for applications in critical care and similar domains.
Authors: Nitish Nagesh, Ziyu Wang, Amir M. Rahmani
Abstract: Synthetic data generation creates data based on real-world data using generative models. In health applications, generating high-quality data while maintaining fairness for sensitive attributes is essential for equitable outcomes. Existing GAN-based and LLM-based methods focus on counterfactual fairness and are primarily applied in finance and legal domains. Causal fairness provides a more comprehensive evaluation framework by preserving causal structure, but current synthetic data generation methods do not address it in health settings. To fill this gap, we develop the first LLM-augmented synthetic data generation method to enhance causal fairness using real-world tabular health data. Our generated data deviates by less than 10% from real data on causal fairness metrics. When trained on causally fair predictors, synthetic data reduces bias on the sensitive attribute by 70% compared to real data. This work improves access to fair synthetic data, supporting equitable health research and healthcare delivery.
Authors: Florian Gr\"otschla, Ahmet Solak, Luca A. Lanzend\"orfer, Roger Wattenhofer
Abstract: Recent advancements have brought generated music closer to human-created compositions, yet evaluating these models remains challenging. While human preference is the gold standard for assessing quality, translating these subjective judgments into objective metrics, particularly for text-audio alignment and music quality, has proven difficult. In this work, we generate 6k songs using 12 state-of-the-art models and conduct a survey of 15k pairwise audio comparisons with 2.5k human participants to evaluate the correlation between human preferences and widely used metrics. To the best of our knowledge, this work is the first to rank current state-of-the-art music generation models and metrics based on human preference. To further the field of subjective metric evaluation, we provide open access to our dataset of generated music and human evaluations.
Authors: Fanny Lehmann, Firat Ozdemir, Benedikt Soja, Torsten Hoefler, Siddhartha Mishra, Sebastian Schemm
Abstract: Recent advances in AI weather forecasting have led to the emergence of so-called "foundation models", typically defined by expensive pretraining and minimal fine-tuning for downstream tasks. However, in the natural sciences, a desirable foundation model should also encode meaningful statistical relationships between the underlying physical variables. This study evaluates the performance of the state-of-the-art Aurora foundation model in predicting hydrological variables, which were not considered during pretraining. We introduce a lightweight approach using shallow decoders trained on the latent representations of the pretrained model to predict these new variables. As a baseline, we compare this to fine-tuning the full model, which allows further optimization of the latent space while incorporating new variables into both inputs and outputs. The decoder-based approach requires 50% less training time and 35% less memory, while achieving strong accuracy across various hydrological variables and preserving desirable properties of the foundation model, such as autoregressive stability. Notably, decoder accuracy depends on the physical correlation between the new variables and those used during pretraining, indicating that Aurora's latent space captures meaningful physical relationships. In this sense, we argue that an important quality metric for foundation models in Earth sciences is their ability to be extended to new variables without a full fine-tuning. This provides a new perspective for making foundation models more accessible to communities with limited computational resources, while supporting broader adoption in Earth sciences.
Authors: Daan Huybrechs
Abstract: It is difficult to describe in mathematical terms what a neural network trained on data represents. On the other hand, there is a growing mathematical understanding of what neural networks are in principle capable of representing. Feedforward neural networks using the ReLU activation function represent continuous and piecewise linear functions and can approximate many others. The study of their expressivity addresses the question: which ones? Contributing to the available answers, we take the perspective of a neural network as an algorithm. In this analogy, a neural network is programmed constructively, rather than trained from data. An interesting example is a sorting algorithm: we explicitly construct a neural network that sorts its inputs exactly, not approximately, and that, in a sense, has optimal computational complexity if the input dimension is large. Such constructed networks may have several billion parameters. We construct and analyze several other examples, both existing and new. We find that, in these examples, neural networks as algorithms are typically recursive and parallel. Compared to conventional algorithms, ReLU networks are restricted by having to be continuous. Moreover, the depth of recursion is limited by the depth of the network, with deep networks having superior properties over shallow ones.
Authors: Kenneth L. Clarkson, Lior Horesh, Takuya Ito, Charlotte Park, Parikshit Ram
Abstract: The invention of the transformer architecture has revolutionized Artificial Intelligence (AI), yielding unprecedented success in areas such as natural language processing, computer vision, and multimodal reasoning. Despite these advances, it is unclear whether transformers are able to learn and implement precise algorithms. Here, we demonstrate that transformers can exactly implement a fundamental and widely used algorithm for $k$-means clustering: Lloyd's algorithm. First, we theoretically prove the existence of such a transformer architecture, which we term the $k$-means transformer, that exactly implements Lloyd's algorithm for $k$-means clustering using the standard ingredients of modern transformers: attention and residual connections. Next, we numerically implement this transformer and demonstrate in experiments the exact correspondence between our architecture and Lloyd's algorithm, providing a fully neural implementation of $k$-means clustering. Finally, we demonstrate that interpretable alterations (e.g., incorporating layer normalizations or multilayer perceptrons) to this architecture yields diverse and novel variants of clustering algorithms, such as soft $k$-means, spherical $k$-means, trimmed $k$-means, and more. Collectively, our findings demonstrate how transformer mechanisms can precisely map onto algorithmic procedures, offering a clear and interpretable perspective on implementing precise algorithms in transformers.
Authors: Andreas Bjerregaard, S{\o}ren Hauberg, Anders Krogh
Abstract: Riemannian representation learning typically relies on approximating densities on chosen manifolds. This involves optimizing difficult objectives, potentially harming models. To completely circumvent this issue, we introduce the Riemannian generative decoder which finds manifold-valued maximum likelihood latents with a Riemannian optimizer while training a decoder network. By discarding the encoder, we vastly simplify the manifold constraint compared to current approaches which often only handle few specific manifolds. We validate our approach on three case studies -- a synthetic branching diffusion process, human migrations inferred from mitochondrial DNA, and cells undergoing a cell division cycle -- each showing that learned representations respect the prescribed geometry and capture intrinsic non-Euclidean structure. Our method requires only a decoder, is compatible with existing architectures, and yields interpretable latent spaces aligned with data geometry.
Authors: Cyrill B\"osch, Geoffrey Roeder, Marc Serra-Garcia, Ryan P. Adams
Abstract: We show that the out-of-equilibrium driving protocol of score-based generative models (SGMs) can be learned via a local learning rule. The gradient with respect to the parameters of the driving protocol are computed directly from force measurements or from observed system dynamics. As a demonstration, we implement an SGM in a network of driven, nonlinear, overdamped oscillators coupled to a thermal bath. We first apply it to the problem of sampling from a mixture of two Gaussians in 2D. Finally, we train a network of 10x10 oscillators to sample images of 0s and 1s from the MNIST dataset.
Authors: Barry Wang, Avi Schwarzschild, Alexander Robey, Ali Payani, Charles Fleming, Mingjie Sun, Daphne Ippolito
Abstract: Retrofitting large language models (LLMs) with new behaviors typically requires full finetuning or distillation-costly steps that must be repeated for every architecture. In this work, we introduce Command-V, a backpropagation-free behavior transfer method that copies an existing residual activation adapter from a donor model and pastes its effect into a recipient model. Command-V profiles layer activations on a small prompt set, derives linear converters between corresponding layers, and applies the donor intervention in the recipient's activation space. This process does not require access to the original training data and needs minimal compute. In three case studies-safety-refusal enhancement, jailbreak facilitation, and automatic chain-of-thought reasoning--Command-V matches or exceeds the performance of direct finetuning while using orders of magnitude less compute. Our code and data are accessible at https://github.com/GithuBarry/Command-V/.
Authors: Paul C. Bogdan, Uzay Macar, Neel Nanda, Arthur Conmy
Abstract: Reasoning large language models have recently achieved state-of-the-art performance in many fields. However, their long-form chain-of-thought reasoning creates interpretability challenges as each generated token depends on all previous ones, making the computation harder to decompose. We argue that analyzing reasoning traces at the sentence level is a promising approach to understanding reasoning processes. We present three complementary attribution methods: (1) a black-box method measuring each sentence's counterfactual importance by comparing final answers across 100 rollouts conditioned on the model generating that sentence or one with a different meaning; (2) a white-box method of aggregating attention patterns between pairs of sentences, which identified "broadcasting" sentences that receive disproportionate attention from all future sentences via "receiver" attention heads; (3) a causal attribution method measuring logical connections between sentences by suppressing attention toward one sentence and measuring the effect on each future sentence's tokens. Each method provides evidence for the existence of thought anchors, reasoning steps that have outsized importance and that disproportionately influence the subsequent reasoning process. These thought anchors are typically planning or backtracking sentences. We provide an open-source tool (www.thought-anchors.com) for visualizing the outputs of our methods, and present a case study showing converging patterns across methods that map how a model performs multi-step reasoning. The consistency across methods demonstrates the potential of sentence-level analysis for a deeper understanding of reasoning models.
Authors: Amir Faiyaz, Tara Salman
Abstract: The rapid proliferation of large language models (LLMs) has created an unprecedented demand for fine-tuning models for specialized domains, such as medical science. While federated learning (FL) offers a decentralized and privacy-preserving approach to collaboratively fine-tune LLMs without sharing raw data, it presents significant challenges, particularly in performance and managing large model sizes efficiently. In this paper, we introduce GradualDiff-Fed, an FL framework designed explicitly for LLMs, and their challenge of handling the high parameter size. GradualDiff-Fed reduces communication costs by transmitting only the difference of model weights rather than the entire model during training rounds. Such an approach significantly improves scalability and communication efficiency, making it more feasible to fine-tune LLMs across distributed clients without compromising performance. Our evaluation demonstrates that GradualDiff-Fed achieves performance on par with centralized training while drastically reducing communication overhead. These results highlight the potential of GradualDiff-Fed as an efficient solution for fine-tuning large models from distributed data in privacy-preserving settings without comprising performance.
Authors: Xingyue Huang, Xianglong Hu, Zifeng Ding, Yuan He, Rishabh, Waleed Alzarooni, Ziyu Ye, Wendong Fan, Bailan He, Haige Bo, Changran Hu, Guohao Li
Abstract: Large language models (LLMs) often struggle with mathematical problems that require exact computation or multi-step algebraic reasoning. Tool-integrated reasoning (TIR) offers a promising solution by leveraging external tools such as code interpreters to ensure correctness, but it introduces inference-time dependencies that hinder scalability and deployment. In this work, we propose a new paradigm for distilling tool knowledge into LLMs purely through natural language. We first construct a Solver Agent that solves math problems by interleaving planning, symbolic tool calls, and reflective reasoning. Then, using a back-translation pipeline powered by multiple LLM-based agents, we convert interleaved TIR traces into natural language reasoning traces. A Translator Agent generates explanations for individual tool calls, while a Rephrase Agent merges them into a fluent and globally coherent narrative. Empirically, we show that fine-tuning a small open-source model on these synthesized traces enables it to internalize both tool knowledge and structured reasoning patterns, yielding gains on competition-level math benchmarks without requiring tool access at inference.
Authors: Conor Snedeker, Xinyu Zhou, Raef Bassily
Abstract: We study model personalization under user-level differential privacy (DP) in the shared representation framework. In this problem, there are $n$ users whose data is statistically heterogeneous, and their optimal parameters share an unknown embedding $U^* \in\mathbb{R}^{d\times k}$ that maps the user parameters in $\mathbb{R}^d$ to low-dimensional representations in $\mathbb{R}^k$, where $k\ll d$. Our goal is to privately recover the shared embedding and the local low-dimensional representations with small excess risk in the federated setting. We propose a private, efficient federated learning algorithm to learn the shared embedding based on the FedRep algorithm in [CHM+21]. Unlike [CHM+21], our algorithm satisfies differential privacy, and our results hold for the case of noisy labels. In contrast to prior work on private model personalization [JRS+21], our utility guarantees hold under a larger class of users' distributions (sub-Gaussian instead of Gaussian distributions). Additionally, in natural parameter regimes, we improve the privacy error term in [JRS+21] by a factor of $\widetilde{O}(dk)$. Next, we consider the binary classification setting. We present an information-theoretic construction to privately learn the shared embedding and derive a margin-based accuracy guarantee that is independent of $d$. Our method utilizes the Johnson-Lindenstrauss transform to reduce the effective dimensions of the shared embedding and the users' data. This result shows that dimension-independent risk bounds are possible in this setting under a margin loss.
Authors: Yixuan Wang, Ziming Liu, Zongyi Li, Anima Anandkumar, Thomas Y. Hou
Abstract: We investigate the high-precision training of Physics-Informed Neural Networks (PINNs) in unbounded domains, with a special focus on applications to singularity formulation in PDEs. We propose a modularized approach and study the choices of neural network ansatz, sampling strategy, and optimization algorithm. When combined with rigorous computer-assisted proofs and PDE analysis, the numerical solutions identified by PINNs, provided they are of high precision, can serve as a powerful tool for studying singularities in PDEs. For 1D Burgers equation, our framework can lead to a solution with very high precision, and for the 2D Boussinesq equation, which is directly related to the singularity formulation in 3D Euler and Navier-Stokes equations, we obtain a solution whose loss is $4$ digits smaller than that obtained in \cite{wang2023asymptotic} with fewer training steps. We also discuss potential directions for pushing towards machine precision for higher-dimensional problems.
Authors: Franziskus Steinert, Salem Said, Cyrus Mostajeran
Abstract: The universality properties of kernels characterize the class of functions that can be approximated in the associated reproducing kernel Hilbert space and are of fundamental importance in the theoretical underpinning of kernel methods in machine learning. In this work, we establish fundamental tools for investigating universality properties of kernels in Riemannian symmetric spaces, thereby extending the study of this important topic to kernels in non-Euclidean domains. Moreover, we use the developed tools to prove the universality of several recent examples from the literature on positive definite kernels defined on Riemannian symmetric spaces, thus providing theoretical justification for their use in applications involving manifold-valued data.
Authors: Renzi Meng, Heyi Wang, Yumeng Sun, Qiyuan Wu, Lian Lian, Renhan Zhang
Abstract: This paper addresses the increasingly prominent problem of anomaly detection in distributed systems. It proposes a detection method based on federated contrastive learning. The goal is to overcome the limitations of traditional centralized approaches in terms of data privacy, node heterogeneity, and anomaly pattern recognition. The proposed method combines the distributed collaborative modeling capabilities of federated learning with the feature discrimination enhancement of contrastive learning. It builds embedding representations on local nodes and constructs positive and negative sample pairs to guide the model in learning a more discriminative feature space. Without exposing raw data, the method optimizes a global model through a federated aggregation strategy. Specifically, the method uses an encoder to represent local behavior data in high-dimensional space. This includes system logs, operational metrics, and system calls. The model is trained using both contrastive loss and classification loss to improve its ability to detect fine-grained anomaly patterns. The method is evaluated under multiple typical attack types. It is also tested in a simulated real-time data stream scenario to examine its responsiveness. Experimental results show that the proposed method outperforms existing approaches across multiple performance metrics. It demonstrates strong detection accuracy and adaptability, effectively addressing complex anomalies in distributed environments. Through careful design of key modules and optimization of the training mechanism, the proposed method achieves a balance between privacy preservation and detection performance. It offers a feasible technical path for intelligent security management in distributed systems.
Authors: Hadi Khalaf, Claudio Mayrink Verdun, Alex Oesterling, Himabindu Lakkaraju, Flavio du Pin Calmon
Abstract: A common paradigm to improve the performance of large language models is optimizing for a reward model. Reward models assign a numerical score to LLM outputs indicating, for example, which response would likely be preferred by a user or is most aligned with safety goals. However, reward models are never perfect. They inevitably function as proxies for complex desiderata such as correctness, helpfulness, and safety. By overoptimizing for a misspecified reward, we can subvert intended alignment goals and reduce overall performance -- a phenomenon commonly referred to as reward hacking. In this work, we characterize reward hacking in inference-time alignment and demonstrate when and how we can mitigate it by hedging on the proxy reward. We study this phenomenon under Best-of-$n$ (BoN) and Soft-Best-of-$n$ (SBoN), and we introduce Best-of-Poisson (BoP) that provides an efficient, near-exact approximation of the optimal reward-KL divergence policy at inference time. We show that the characteristic pattern of hacking as observed in practice (where the true reward first increases before declining) is an inevitable property of a broad class of inference-time mechanisms, including BoN and BoP. To counter this effect, hedging offers a tactical choice to avoid placing undue confidence in high but potentially misleading proxy reward signals. We introduce HedgeTune, an efficient algorithm to find the optimal inference-time parameter and avoid reward hacking. We demonstrate through experiments that hedging mitigates reward hacking and achieves superior distortion-reward tradeoffs with minimal computational overhead.
Authors: Shili Wu, Yizhao Jin, Puhua Niu, Aniruddha Datta, Sean B. Andersson
Abstract: Behavior Cloning (BC) is an effective imitation learning technique and has even been adopted in some safety-critical domains such as autonomous vehicles. BC trains a policy to mimic the behavior of an expert by using a dataset composed of only state-action pairs demonstrated by the expert, without any additional interaction with the environment. However, During deployment, the policy observations may contain measurement errors or adversarial disturbances. Since the observations may deviate from the true states, they can mislead the agent into making sub-optimal actions. In this work, we use a global Lipschitz regularization approach to enhance the robustness of the learned policy network. We then show that the resulting global Lipschitz property provides a robustness certificate to the policy with respect to different bounded norm perturbations. Then, we propose a way to construct a Lipschitz neural network that ensures the policy robustness. We empirically validate our theory across various environments in Gymnasium. Keywords: Robust Reinforcement Learning; Behavior Cloning; Lipschitz Neural Network
Authors: Yang Zhou, Xiaoning Ren
Abstract: Graph Out-of-Distribution (OOD) classification often suffers from sharp performance drops, particularly under category imbalance and structural noise. This work tackles two pressing challenges in this context: (1) the underperformance of minority classes due to skewed label distributions, and (2) their heightened sensitivity to structural noise in graph data. To address these problems, we propose two complementary solutions. First, Constrained Mean Optimization (CMO) improves minority class robustness by encouraging similarity-based instance aggregation under worst-case conditions. Second, the Neighbor-Aware Noise Reweighting (NNR) mechanism assigns dynamic weights to training samples based on local structural consistency, mitigating noise influence. We provide theoretical justification for our methods, and validate their effectiveness with extensive experiments on both synthetic and real-world datasets, showing significant improvements in Graph OOD generalization and classification accuracy. The code for our method is available at: https://anonymous.4open.science/r/CMO-NNR-2F30.
Authors: Yang Zhou, Xiaoning Ren
Abstract: In dynamic graphs, preserving temporal continuity is critical. However, Memory-based Dynamic Graph Neural Networks (MDGNNs) trained with large batches often disrupt event sequences, leading to temporal information loss. This discontinuity not only deteriorates temporal modeling but also hinders optimization by increasing the difficulty of parameter convergence. Our theoretical study quantifies this through a Lipschitz upper bound, showing that large batch sizes enlarge the parameter search space. In response, we propose BADGNN, a novel batch-agnostic framework consisting of two core components: (1) Temporal Lipschitz Regularization (TLR) to control parameter search space expansion, and (2) Adaptive Attention Adjustment (A3) to alleviate attention distortion induced by both regularization and batching. Empirical results on three benchmark datasets show that BADGNN maintains strong performance while enabling significantly larger batch sizes and faster training compared to TGN. Our code is available at Code: https://anonymous.4open.science/r/TGN_Lipichitz-C033/.
URLs: https://anonymous.4open.science/r/TGN_Lipichitz-C033/.
Authors: Yan Li, Zengli Yang, Youhuai Wang, Jing Wang, Xiaoyu Han, Jingyu Wang, Dongyuan Shi
Abstract: The Extreme Operating Conditions Search (EOCS) problem is one of the key problems in relay setting calculation, which is used to ensure that the setting values of protection relays can adapt to the changing operating conditions of power systems over a period of time after deployment. The high penetration of renewable energy and the wide application of inverter-based resources make the operating conditions of renewable power systems more volatile, which urges the adoption of the online relay setting calculation strategy. However, the computation speed of existing EOCS methods based on local enumeration, heuristic algorithms, and mathematical programming cannot meet the efficiency requirement of online relay setting calculation. To reduce the time overhead, this paper, for the first time, proposes an efficient deep learning-based EOCS method suitable for online relay setting calculation. First, the power system information is formulated as four layers, i.e., a component parameter layer, a topological connection layer, an electrical distance layer, and a graph distance layer, which are fed into a parallel graph neural network (PGNN) model for feature extraction. Then, the four feature layers corresponding to each node are spliced and stretched, and then fed into the decision network to predict the extreme operating condition of the system. Finally, the proposed PGNN method is validated on the modified IEEE 39-bus and 118-bus test systems, where some of the synchronous generators are replaced by renewable generation units. The nonlinear fault characteristics of renewables are fully considered when computing fault currents. The experiment results show that the proposed PGNN method achieves higher accuracy than the existing methods in solving the EOCS problem. Meanwhile, it also provides greater improvements in online computation time.
Authors: Zeyu Bao, Penghao Yu, Haotian Jiang, Qianxiao Li
Abstract: Deep state-space models (SSMs) have gained increasing popularity in sequence modelling. While there are numerous theoretical investigations of shallow SSMs, how the depth of the SSM affects its expressiveness remains a crucial problem. In this paper, we systematically investigate the role of depth and width in deep linear SSMs, aiming to characterize how they influence the expressive capacity of the architecture. First, we rigorously prove that in the absence of parameter constraints, increasing depth and increasing width are generally equivalent, provided that the parameter count remains within the same order of magnitude. However, under the assumption that the parameter norms are constrained, the effects of depth and width differ significantly. We show that a shallow linear SSM with large parameter norms can be represented by a deep linear SSM with smaller norms using a constructive method. In particular, this demonstrates that deep SSMs are more capable of representing targets with large norms than shallow SSMs under norm constraints. Finally, we derive upper bounds on the minimal depth required for a deep linear SSM to represent a given shallow linear SSM under constrained parameter norms. We also validate our theoretical results with numerical experiments
Authors: Ahmad Mohammad Saber, Aditi Maheshwari, Amr Youssef, Deepa Kundur
Abstract: The application of Deep Learning-based Schemes (DLSs) for detecting False Data Injection Attacks (FDIAs) in smart grids has attracted significant attention. This paper demonstrates that adversarial attacks, carefully crafted FDIAs, can evade existing DLSs used for FDIA detection in Line Current Differential Relays (LCDRs). We propose a novel adversarial attack framework, utilizing the Fast Gradient Sign Method, which exploits DLS vulnerabilities by introducing small perturbations to LCDR remote measurements, leading to misclassification of the FDIA as a legitimate fault while also triggering the LCDR to trip. We evaluate the robustness of multiple deep learning models, including multi-layer perceptrons, convolutional neural networks, long short-term memory networks, and residual networks, under adversarial conditions. Our experimental results demonstrate that while these models perform well, they exhibit high degrees of vulnerability to adversarial attacks. For some models, the adversarial attack success rate exceeds 99.7%. To address this threat, we introduce adversarial training as a proactive defense mechanism, significantly enhancing the models' ability to withstand adversarial FDIAs without compromising fault detection accuracy. Our results highlight the significant threat posed by adversarial attacks to DLS-based FDIA detection, underscore the necessity for robust cybersecurity measures in smart grids, and demonstrate the effectiveness of adversarial training in enhancing model robustness against adversarial FDIAs.
Authors: Vineet Punyamoorty, Aditya Malusare, Vaneet Aggarwal
Abstract: Modern diagnostic workflows are increasingly multimodal, integrating diverse data sources such as medical images, structured records, and physiological time series. Among these, electrocardiograms (ECGs) and chest X-rays (CXRs) are two of the most widely used modalities for cardiac assessment. While CXRs provide rich diagnostic information, ECGs are more accessible and can support scalable early warning systems. In this work, we propose CroMoTEX, a novel contrastive learning-based framework that leverages chest X-rays during training to learn clinically informative ECG representations for multiple cardiac-related pathologies: cardiomegaly, pleural effusion, and edema. Our method aligns ECG and CXR representations using a novel supervised cross-modal contrastive objective with adaptive hard negative weighting, enabling robust and task-relevant feature learning. At test time, CroMoTEX relies solely on ECG input, allowing scalable deployment in real-world settings where CXRs may be unavailable. Evaluated on the large-scale MIMIC-IV-ECG and MIMIC-CXR datasets, CroMoTEX outperforms baselines across all three pathologies, achieving up to 78.31 AUROC on edema. Our code is available at github.com/vineetpmoorty/cromotex.
Authors: Sudesh Bhagat, Raghupathi Kandiboina, Ibne Farabi Shihab, Skylar Knickerbocker, Neal Hawkins, Anuj Sharma
Abstract: Road traffic crashes are a significant global cause of fatalities, emphasizing the urgent need for accurate crash data to enhance prevention strategies and inform policy development. This study addresses the challenge of alcohol inference mismatch (AIM) by employing database narrative alignment to identify AIM in crash data. A framework was developed to improve data quality in crash management systems and reduce the percentage of AIM crashes. Utilizing the BERT model, the analysis of 371,062 crash records from Iowa (2016-2022) revealed 2,767 AIM incidents, resulting in an overall AIM percentage of 24.03%. Statistical tools, including the Probit Logit model, were used to explore the crash characteristics affecting AIM patterns. The findings indicate that alcohol-related fatal crashes and nighttime incidents have a lower percentage of the mismatch, while crashes involving unknown vehicle types and older drivers are more susceptible to mismatch. The geospatial cluster as part of this study can identify the regions which have an increased need for education and training. These insights highlight the necessity for targeted training programs and data management teams to improve the accuracy of crash reporting and support evidence-based policymaking.
Authors: Ziyu Zheng, Yaming Yang, Ziyu Guan, Wei Zhao, Weigang Lu
Abstract: Masked Graph Auto-Encoder, a powerful graph self-supervised training paradigm, has recently shown superior performance in graph representation learning. Existing works typically rely on node contextual information to recover the masked information. However, they fail to generalize well to heterophilic graphs where connected nodes may be not similar, because they focus only on capturing the neighborhood information and ignoring the discrepancy information between different nodes, resulting in indistinguishable node representations. In this paper, to address this issue, we propose a Discrepancy-Aware Graph Mask Auto-Encoder (DGMAE). It obtains more distinguishable node representations by reconstructing the discrepancy information of neighboring nodes during the masking process. We conduct extensive experiments on 17 widely-used benchmark datasets. The results show that our DGMAE can effectively preserve the discrepancies of nodes in low-dimensional space. Moreover, DGMAE significantly outperforms state-of-the-art graph self-supervised learning methods on three graph analytic including tasks node classification, node clustering, and graph classification, demonstrating its remarkable superiority. The code of DGMAE is available at https://github.com/zhengziyu77/DGMAE.
Authors: Puneesh Deora, Bhavya Vasudeva, Tina Behnia, Christos Thrampoulidis
Abstract: In-context learning (ICL) enables transformers to adapt to new tasks through contextual examples without parameter updates. While existing research has typically studied ICL in fixed-complexity environments, practical language models encounter tasks spanning diverse complexity levels. This paper investigates how transformers navigate hierarchical task structures where higher-complexity categories can perfectly represent any pattern generated by simpler ones. We design well-controlled testbeds based on Markov chains and linear regression that reveal transformers not only identify the appropriate complexity level for each task but also accurately infer the corresponding parameters--even when the in-context examples are compatible with multiple complexity hypotheses. Notably, when presented with data generated by simpler processes, transformers consistently favor the least complex sufficient explanation. We theoretically explain this behavior through a Bayesian framework, demonstrating that transformers effectively implement an in-context Bayesian Occam's razor by balancing model fit against complexity penalties. We further ablate on the roles of model size, training mixture distribution, inference context length, and architecture. Finally, we validate this Occam's razor-like inductive bias on a pretrained GPT-4 model with Boolean-function tasks as case study, suggesting it may be inherent to transformers trained on diverse task distributions.
Authors: Kohei Miyaguchi
Abstract: In this paper, we propose trajectory advantage regression, a method of offline path learning and path attribution based on reinforcement learning. The proposed method can be used to solve path optimization problems while algorithmically only solving a regression problem.
Authors: Shreya, Harsh Pathak
Abstract: This paper presents an intelligent and transparent AI-driven system for Credit Risk Assessment using three state-of-the-art ensemble machine learning models combined with Explainable AI (XAI) techniques. The system leverages XGBoost, LightGBM, and Random Forest algorithms for predictive analysis of loan default risks, addressing the challenges of model interpretability using SHAP and LIME. Preprocessing steps include custom imputation, one-hot encoding, and standardization. Class imbalance is managed using SMOTE, and hyperparameter tuning is performed with GridSearchCV. The model is evaluated on multiple performance metrics including ROC-AUC, precision, recall, and F1-score. LightGBM emerges as the most business-optimal model with the highest accuracy and best trade off between approval and default rates. Furthermore, the system generates applicant-specific XAI visual reports and business impact summaries to ensure transparent decision-making.
Authors: Shijian Zheng, Fangxiao Jin, Shuhai Zhang, Quan Xue, Mingkui Tan
Abstract: Electromagnetic structure (EMS) design plays a critical role in developing advanced antennas and materials, but remains challenging due to high-dimensional design spaces and expensive evaluations. While existing methods commonly employ high-quality predictors or generators to alleviate evaluations, they are often data-intensive and struggle with real-world scale and budget constraints. To address this, we propose a novel method called Progressive Quadtree-based Search (PQS). Rather than exhaustively exploring the high-dimensional space, PQS converts the conventional image-like layout into a quadtree-based hierarchical representation, enabling a progressive search from global patterns to local details. Furthermore, to lessen reliance on highly accurate predictors, we introduce a consistency-driven sample selection mechanism. This mechanism quantifies the reliability of predictions, balancing exploitation and exploration when selecting candidate designs. We evaluate PQS on two real-world engineering tasks, i.e., Dual-layer Frequency Selective Surface and High-gain Antenna. Experimental results show that our method can achieve satisfactory designs under limited computational budgets, outperforming baseline methods. In particular, compared to generative approaches, it cuts evaluation costs by 75-85%, effectively saving 20.27-38.80 days of product designing cycle.
Authors: Shanda Li, Shinjae Yoo, Yiming Yang
Abstract: Fourier Neural Operators (FNOs) offer a principled approach for solving complex partial differential equations (PDEs). However, scaling them to handle more complex PDEs requires increasing the number of Fourier modes, which significantly expands the number of model parameters and makes hyperparameter tuning computationally impractical. To address this, we introduce $\mu$Transfer-FNO, a zero-shot hyperparameter transfer technique that enables optimal configurations, tuned on smaller FNOs, to be directly applied to billion-parameter FNOs without additional tuning. Building on the Maximal Update Parametrization ($\mu$P) framework, we mathematically derive a parametrization scheme that facilitates the transfer of optimal hyperparameters across models with different numbers of Fourier modes in FNOs, which is validated through extensive experiments on various PDEs. Our empirical study shows that Transfer-FNO reduces computational cost for tuning hyperparameters on large FNOs while maintaining or improving accuracy.
Authors: Yisak Park, Sunwoo Lee, Seungyul Han
Abstract: Cooperative multi-agent reinforcement learning (MARL) under sparse rewards presents a fundamental challenge due to limited exploration and insufficient coordinated attention among agents. In this work, we propose the Focusing Influence Mechanism (FIM), a novel framework that enhances cooperation by directing agent influence toward task-critical elements, referred to as Center of Gravity (CoG) state dimensions, inspired by Clausewitz's military theory. FIM consists of three core components: (1) identifying CoG state dimensions based on their stability under agent behavior, (2) designing counterfactual intrinsic rewards to promote meaningful influence on these dimensions, and (3) encouraging persistent and synchronized focus through eligibility-trace-based credit accumulation. These mechanisms enable agents to induce more targeted and effective state transitions, facilitating robust cooperation even in extremely sparse reward settings. Empirical evaluations across diverse MARL benchmarks demonstrate that the proposed FIM significantly improves cooperative performance compared to baselines.
Authors: Florian Peter Busch, Moritz Willig, Florian Guldan, Kristian Kersting, Devendra Singh Dhami
Abstract: Not every causal relation between variables is equal, and this can be leveraged for the task of causal discovery. Recent research shows that pairs of variables with particular type assignments induce a preference on the causal direction of other pairs of variables with the same type. Although useful, this assignment of a specific type to a variable can be tricky in practice. We propose a tag-based causal discovery approach where multiple tags are assigned to each variable in a causal graph. Existing causal discovery approaches are first applied to direct some edges, which are then used to determine edge relations between tags. Then, these edge relations are used to direct the undirected edges. Doing so improves upon purely type-based relations, where the assumption of type consistency lacks robustness and flexibility due to being restricted to single types for each variable. Our experimental evaluations show that this boosts causal discovery and that these high-level tag relations fit common knowledge.
Authors: Leif D\"oring, Benedikt Wille, Maximilian Birr, Mihail B\^irsan, Martin Slowik
Abstract: Bias problems in the estimation of $Q$-values are a well-known obstacle that slows down convergence of $Q$-learning and actor-critic methods. One of the reasons of the success of modern RL algorithms is partially a direct or indirect overestimation reduction mechanism. We propose an easy to implement method built on top of distributional reinforcement learning (DRL) algorithms to deal with the overestimation in a locally adaptive way. Our framework is simple to implement, existing distributional algorithms can be improved with a few lines of code. We provide theoretical evidence and use double $Q$-learning to show how to include locally adaptive overestimation control in existing algorithms. Experiments are provided for tabular, Atari, and MuJoCo environments.
Authors: Yuelin Zhang, Jiacheng Cen, Jiaqi Han, Wenbing Huang
Abstract: Equivariant Graph Neural Networks (GNNs) have achieved remarkable success across diverse scientific applications. However, existing approaches face critical efficiency challenges when scaling to large geometric graphs and suffer significant performance degradation when the input graphs are sparsified for computational tractability. To address these limitations, we introduce FastEGNN and DistEGNN, two novel enhancements to equivariant GNNs for large-scale geometric graphs. FastEGNN employs a key innovation: a small ordered set of virtual nodes that effectively approximates the large unordered graph of real nodes. Specifically, we implement distinct message passing and aggregation mechanisms for different virtual nodes to ensure mutual distinctiveness, and minimize Maximum Mean Discrepancy (MMD) between virtual and real coordinates to achieve global distributedness. This design enables FastEGNN to maintain high accuracy while efficiently processing large-scale sparse graphs. For extremely large-scale geometric graphs, we present DistEGNN, a distributed extension where virtual nodes act as global bridges between subgraphs in different devices, maintaining consistency while dramatically reducing memory and computational overhead. We comprehensively evaluate our models across four challenging domains: N-body systems (100 nodes), protein dynamics (800 nodes), Water-3D (8,000 nodes), and our new Fluid113K benchmark (113,000 nodes). Results demonstrate superior efficiency and performance, establishing new capabilities in large-scale equivariant graph learning. Code is available at https://github.com/GLAD-RUC/DistEGNN.
Authors: Zhihao Sui, Liang Hu, Jian Cao, Dora D. Liu, Usman Naseem, Zhongyuan Lai, Qi Zhang
Abstract: Machine Unlearning (MU) technology facilitates the removal of the influence of specific data instances from trained models on request. Despite rapid advancements in MU technology, its vulnerabilities are still underexplored, posing potential risks of privacy breaches through leaks of ostensibly unlearned information. Current limited research on MU attacks requires access to original models containing privacy data, which violates the critical privacy-preserving objective of MU. To address this gap, we initiate an innovative study on recalling the forgotten class memberships from unlearned models (ULMs) without requiring access to the original one. Specifically, we implement a Membership Recall Attack (MRA) framework with a teacher-student knowledge distillation architecture, where ULMs serve as noisy labelers to transfer knowledge to student models. Then, it is translated into a Learning with Noisy Labels (LNL) problem for inferring the correct labels of the forgetting instances. Extensive experiments on state-of-the-art MU methods with multiple real datasets demonstrate that the proposed MRA strategy exhibits high efficacy in recovering class memberships of unlearned instances. As a result, our study and evaluation have established a benchmark for future research on MU vulnerabilities.
Authors: Zhihao Sui, Liang Hu, Jian Cao, Usman Naseem, Zhongyuan Lai, Qi Zhang
Abstract: Large deep learning models have achieved significant success in various tasks. However, the performance of a model can significantly degrade if it is needed to train on datasets with noisy labels with misleading or ambiguous information. To date, there are limited investigations on how to restore performance when model degradation has been incurred by noisy label data. Inspired by the ``forgetting mechanism'' in neuroscience, which enables accelerating the relearning of correct knowledge by unlearning the wrong knowledge, we propose a robust model restoration and refinement (MRR) framework COLUR, namely Confidence-Oriented Learning, Unlearning and Relearning. Specifically, we implement COLUR with an efficient co-training architecture to unlearn the influence of label noise, and then refine model confidence on each label for relearning. Extensive experiments are conducted on four real datasets and all evaluation results show that COLUR consistently outperforms other SOTA methods after MRR.
Authors: Paul Kahlmeyer, Markus Fischer, Joachim Giesen
Abstract: Solutions of symbolic regression problems are expressions that are composed of input variables and operators from a finite set of function symbols. One measure for evaluating symbolic regression algorithms is their ability to recover formulae, up to symbolic equivalence, from finite samples. Not unexpectedly, the recovery problem becomes harder when the formula gets more complex, that is, when the number of variables and operators gets larger. Variables in naturally occurring symbolic formulas often appear only in fixed combinations. This can be exploited in symbolic regression by substituting one new variable for the combination, effectively reducing the number of variables. However, finding valid substitutions is challenging. Here, we address this challenge by searching over the expression space of small substitutions and testing for validity. The validity test is reduced to a test of functional dependence. The resulting iterative dimension reduction procedure can be used with any symbolic regression approach. We show that it reliably identifies valid substitutions and significantly boosts the performance of different types of state-of-the-art symbolic regression algorithms.
Authors: Lennart Schneider, Bernd Bischl, Matthias Feurer
Abstract: Hyperparameter optimization (HPO) aims to identify an optimal hyperparameter configuration (HPC) such that the resulting model generalizes well to unseen data. As the expected generalization error cannot be optimized directly, it is estimated with a resampling strategy, such as holdout or cross-validation. This approach implicitly assumes that minimizing the validation error leads to improved generalization. However, since validation error estimates are inherently stochastic and depend on the resampling strategy, a natural question arises: Can excessive optimization of the validation error lead to overfitting at the HPO level, akin to overfitting in model training based on empirical risk minimization? In this paper, we investigate this phenomenon, which we term overtuning, a form of overfitting specific to HPO. Despite its practical relevance, overtuning has received limited attention in the HPO and AutoML literature. We provide a formal definition of overtuning and distinguish it from related concepts such as meta-overfitting. We then conduct a large-scale reanalysis of HPO benchmark data to assess the prevalence and severity of overtuning. Our results show that overtuning is more common than previously assumed, typically mild but occasionally severe. In approximately 10% of cases, overtuning leads to the selection of a seemingly optimal HPC with worse generalization error than the default or first configuration tried. We further analyze how factors such as performance metric, resampling strategy, dataset size, learning algorithm, and HPO method affect overtuning and discuss mitigation strategies. Our results highlight the need to raise awareness of overtuning, particularly in the small-data regime, indicating that further mitigation strategies should be studied.
Authors: Paul Kahlmeyer, Niklas Merk, Joachim Giesen
Abstract: Solving systems of ordinary differential equations (ODEs) is essential when it comes to understanding the behavior of dynamical systems. Yet, automated solving remains challenging, in particular for nonlinear systems. Computer algebra systems (CASs) provide support for solving ODEs by first simplifying them, in particular through the use of Lie point symmetries. Finding these symmetries is, however, itself a difficult problem for CASs. Recent works in symbolic regression have shown promising results for recovering symbolic expressions from data. Here, we adapt search-based symbolic regression to the task of finding generators of Lie point symmetries. With this approach, we can find symmetries of ODEs that existing CASs cannot find.
Authors: QinZhe Wang, Zixuan Chen, Keke Huang, Xiu Su, Chunhua Yang, Chang Xu
Abstract: Few-Shot Class-Incremental Learning (FSCIL) requires models to adapt to novel classes with limited supervision while preserving learned knowledge. Existing prospective learning-based space construction methods reserve space to accommodate novel classes. However, prototype deviation and structure fixity limit the expressiveness of the embedding space. In contrast to fixed space reservation, we explore the optimization of feature-structure dual consistency and propose a Consistency-driven Calibration and Matching Framework (ConCM) that systematically mitigate the knowledge conflict inherent in FSCIL. Specifically, inspired by hippocampal associative memory, we design a memory-aware prototype calibration that extracts generalized semantic attributes from base classes and reintegrates them into novel classes to enhance the conceptual center consistency of features. Further, we propose dynamic structure matching, which adaptively aligns the calibrated features to a session-specific optimal manifold space, ensuring cross-session structure consistency. Theoretical analysis shows that our method satisfies both geometric optimality and maximum matching, thereby overcoming the need for class-number priors. On large-scale FSCIL benchmarks including mini-ImageNet and CUB200, ConCM achieves state-of-the-art performance, surpassing current optimal method by 3.20% and 3.68% in harmonic accuracy of incremental sessions.
Authors: Pengpeng Ouyang, Dong Chen, Tong Yang, Shuo Feng, Zhao Jin, Mingliang Xu
Abstract: Multi-task and few-shot time series forecasting tasks are commonly encountered in scenarios such as the launch of new products in different cities. However, traditional time series forecasting methods suffer from insufficient historical data, which stems from a disregard for the generalized and specific features among different tasks. For the aforementioned challenges, we propose the Feature-Adaptive Time Series Forecasting Framework (FAF), which consists of three key components: the Generalized Knowledge Module (GKM), the Task-Specific Module (TSM), and the Rank Module (RM). During training phase, the GKM is updated through a meta-learning mechanism that enables the model to extract generalized features across related tasks. Meanwhile, the TSM is trained to capture diverse local dynamics through multiple functional regions, each of which learns specific features from individual tasks. During testing phase, the RM dynamically selects the most relevant functional region from the TSM based on input sequence features, which is then combined with the generalized knowledge learned by the GKM to generate accurate forecasts. This design enables FAF to achieve robust and personalized forecasting even with sparse historical observations We evaluate FAF on five diverse real-world datasets under few-shot time series forecasting settings. Experimental results demonstrate that FAF consistently outperforms baselines that include three categories of time series forecasting methods. In particular, FAF achieves a 41.81\% improvement over the best baseline, iTransformer, on the CO$_2$ emissions dataset.
Authors: Santiago A. Cadena, Andrea Merlo, Emanuel Laude, Alexander Bauer, Atul Agrawal, Maria Pascu, Marija Savtchouk, Enrico Guiraud, Lukas Bonauer, Stuart Hudson, Markus Kaiser
Abstract: Stellarators are magnetic confinement devices under active development to deliver steady-state carbon-free fusion energy. Their design involves a high-dimensional, constrained optimization problem that requires expensive physics simulations and significant domain expertise. Recent advances in plasma physics and open-source tools have made stellarator optimization more accessible. However, broader community progress is currently bottlenecked by the lack of standardized optimization problems with strong baselines and datasets that enable data-driven approaches, particularly for quasi-isodynamic (QI) stellarator configurations, considered as a promising path to commercial fusion due to their inherent resilience to current-driven disruptions. Here, we release an open dataset of diverse QI-like stellarator plasma boundary shapes, paired with their ideal magnetohydrodynamic (MHD) equilibria and performance metrics. We generated this dataset by sampling a variety of QI fields and optimizing corresponding stellarator plasma boundaries. We introduce three optimization benchmarks of increasing complexity: (1) a single-objective geometric optimization problem, (2) a "simple-to-build" QI stellarator, and (3) a multi-objective ideal-MHD stable QI stellarator that investigates trade-offs between compactness and coil simplicity. For every benchmark, we provide reference code, evaluation scripts, and strong baselines based on classical optimization techniques. Finally, we show how learned models trained on our dataset can efficiently generate novel, feasible configurations without querying expensive physics oracles. By openly releasing the dataset along with benchmark problems and baselines, we aim to lower the entry barrier for optimization and machine learning researchers to engage in stellarator design and to accelerate cross-disciplinary progress toward bringing fusion energy to the grid.
Authors: Alan N. Amin, Andres Potapczynski, Andrew Gordon Wilson
Abstract: To understand how genetic variants in human genomes manifest in phenotypes -- traits like height or diseases like asthma -- geneticists have sequenced and measured hundreds of thousands of individuals. Geneticists use this data to build models that predict how a genetic variant impacts phenotype given genomic features of the variant, like DNA accessibility or the presence of nearby DNA-bound proteins. As more data and features become available, one might expect predictive models to improve. Unfortunately, training these models is bottlenecked by the need to solve expensive linear algebra problems because variants in the genome are correlated with nearby variants, requiring inversion of large matrices. Previous methods have therefore been restricted to fitting small models, and fitting simplified summary statistics, rather than the full likelihood of the statistical model. In this paper, we leverage modern fast linear algebra techniques to develop DeepWAS (Deep genome Wide Association Studies), a method to train large and flexible neural network predictive models to optimize likelihood. Notably, we find that larger models only improve performance when using our full likelihood approach; when trained by fitting traditional summary statistics, larger models perform no better than small ones. We find larger models trained on more features make better predictions, potentially improving disease predictions and therapeutic target identification.
Authors: Pantelis R. Vlachas, Konstantinos Vlachas, Eleni Chatzi
Abstract: Dynamical systems play a key role in modeling, forecasting, and decision-making across a wide range of scientific domains. However, variations in system parameters, also referred to as parametric variability, can lead to drastically different model behavior and output, posing challenges for constructing models that generalize across parameter regimes. In this work, we introduce the Parametric Hypernetwork for Learning Interpolated Networks (PHLieNet), a framework that simultaneously learns: (a) a global mapping from the parameter space to a nonlinear embedding and (b) a mapping from the inferred embedding to the weights of a dynamics propagation network. The learned embedding serves as a latent representation that modulates a base network, termed the hypernetwork, enabling it to generate the weights of a target network responsible for forecasting the system's state evolution conditioned on the previous time history. By interpolating in the space of models rather than observations, PHLieNet facilitates smooth transitions across parameterized system behaviors, enabling a unified model that captures the dynamic behavior across a broad range of system parameterizations. The performance of the proposed technique is validated in a series of dynamical systems with respect to its ability to extrapolate in time and interpolate and extrapolate in the parameter space, i.e., generalize to dynamics that were unseen during training. In all cases, our approach outperforms or matches state-of-the-art baselines in both short-term forecast accuracy and in capturing long-term dynamical features, such as attractor statistics.
Authors: Paul Kahlmeyer, Joachim Giesen, Michael Habeck, Henrik Voigt
Abstract: In a regression task, a function is learned from labeled data to predict the labels at new data points. The goal is to achieve small prediction errors. In symbolic regression, the goal is more ambitious, namely, to learn an interpretable function that makes small prediction errors. This additional goal largely rules out the standard approach used in regression, that is, reducing the learning problem to learning parameters of an expansion of basis functions by optimization. Instead, symbolic regression methods search for a good solution in a space of symbolic expressions. To cope with the typically vast search space, most symbolic regression methods make implicit, or sometimes even explicit, assumptions about its structure. Here, we argue that the only obvious structure of the search space is that it contains small expressions, that is, expressions that can be decomposed into a few subexpressions. We show that systematically searching spaces of small expressions finds solutions that are more accurate and more robust against noise than those obtained by state-of-the-art symbolic regression methods. In particular, systematic search outperforms state-of-the-art symbolic regressors in terms of its ability to recover the true underlying symbolic expressions on established benchmark data sets.
Authors: Thomas Decker, Volker Tresp, Florian Buettner
Abstract: Perturbation-based explanations are widely utilized to enhance the transparency of modern machine-learning models. However, their reliability is often compromised by the unknown model behavior under the specific perturbations used. This paper investigates the relationship between uncertainty calibration - the alignment of model confidence with actual accuracy - and perturbation-based explanations. We show that models frequently produce unreliable probability estimates when subjected to explainability-specific perturbations and theoretically prove that this directly undermines explanation quality. To address this, we introduce ReCalX, a novel approach to recalibrate models for improved perturbation-based explanations while preserving their original predictions. Experiments on popular computer vision models demonstrate that our calibration strategy produces explanations that are more aligned with human perception and actual object locations.
Authors: Alessandro Salatiello, Stefan Birr, Manuel Kunz
Abstract: Coherently forecasting the behaviour of a target variable across both coarse and fine temporal scales is crucial for profit-optimized decision-making in several business applications, and remains an open research problem in temporal hierarchical forecasting. Here, we propose a new hierarchical architecture that tackles this problem by leveraging modules that specialize in forecasting the different temporal aggregation levels of interest. The architecture, which learns to encode the average behaviour of the target variable within its hidden layers, makes accurate and coherent forecasts across the target temporal hierarchies. We validate our architecture on the challenging, real-world M5 dataset and show that it outperforms established methods, such as the TSMixer model.
Authors: Shuncheng He, Hongchang Zhang, Jianzhun Shao, Yuhang Jiang, Xiangyang Ji
Abstract: Offline reinforcement learning (RL) recently gains growing interests from RL researchers. However, the performance of offline RL suffers from the out-of-distribution problem, which can be corrected by feedback in online RL. Previous offline RL research focuses on restricting the offline algorithm in in-distribution even in-sample action sampling. In contrast, fewer work pays attention to the influence of the batch data. In this paper, we first build a bridge over the batch data and the performance of offline RL algorithms theoretically, from the perspective of model-based offline RL optimization. We draw a conclusion that, with mild assumptions, the distance between the state-action pair distribution generated by the behavioural policy and the distribution generated by the optimal policy, accounts for the performance gap between the policy learned by model-based offline RL and the optimal policy. Secondly, we reveal that in task-agnostic settings, a series of policies trained by unsupervised RL can minimize the worst-case regret in the performance gap. Inspired by the theoretical conclusions, UDG (Unsupervised Data Generation) is devised to generate data and select proper data for offline training under tasks-agnostic settings. Empirical results demonstrate that UDG can outperform supervised data generation on solving unknown tasks.
Authors: Itay Lamprecht, Asaf Karnieli, Yair Hanani, Niv Giladi, Daniel Soudry
Abstract: Training and inference of Large Language Models (LLMs) with tensor-parallelism requires substantial communication to synchronize activations. Our findings suggest that with a few minor adjustments to current practices, LLMs can be trained without fully synchronizing activations, reducing bandwidth demands. We name this "Communication-Aware Architecture for Tensor-parallelism" (CAAT-Net). We train 1B and 7B parameter CAAT-Net models, with a 50% reduction in tensor-parallel communication and no significant drop in pretraining accuracy. Furthermore, we demonstrate how CAAT-Net accelerates both training and inference workloads.
Authors: Mihnea Ghitu, Matthew Wicker, Vihari Piratla
Abstract: Controlling the patterns a model learns is essential to preventing reliance on irrelevant or misleading features. Such reliance on irrelevant features, often called shortcut features, has been observed across domains, including medical imaging and natural language processing, where it may lead to real-world harms. A common mitigation strategy leverages annotations (provided by humans or machines) indicating which features are relevant or irrelevant. These annotations are compared to model explanations, typically in the form of feature salience, and used to guide the loss function during training. Unfortunately, recent works have demonstrated that feature salience methods are unreliable and therefore offer a poor signal to optimize. In this work, we propose a simplified objective that simultaneously optimizes for explanation robustness and mitigation of shortcut learning. Unlike prior objectives with similar aims, we demonstrate theoretically why our approach ought to be more effective. Across a comprehensive series of experiments, we show that our approach consistently reduces test-time misclassifications by 20% compared to state-of-the-art methods. We also extend prior experimental settings to include natural language processing tasks. Additionally, we conduct novel ablations that yield practical insights, including the relative importance of annotation quality over quantity. Code for our method and experiments is available at: https://github.com/Mihneaghitu/ModelGuidanceViaRobustFeatureAttribution.
URLs: https://github.com/Mihneaghitu/ModelGuidanceViaRobustFeatureAttribution.
Authors: A. A. Balinsky, A. D. Balinsky
Abstract: Reliable uncertainty quantification is crucial for the trustworthiness of machine learning applications. Inductive Conformal Prediction (ICP) offers a distribution-free framework for generating prediction sets or intervals with user-specified confidence. However, standard ICP guarantees are marginal and typically require a fresh calibration set for each new prediction to maintain their validity. This paper addresses this practical limitation by demonstrating how e-conformal prediction, in conjunction with Hoeffding's inequality, can enable the repeated use of a single calibration set with a high probability of preserving the desired coverage. Through a case study on the CIFAR-10 dataset, we train a deep neural network and utilise a calibration set to estimate a Hoeffding correction. This correction allows us to apply a modified Markov's inequality, leading to the construction of prediction sets with quantifiable confidence. Our results illustrate the feasibility of maintaining provable performance in conformal prediction while enhancing its practicality by reducing the need for repeated calibration. The code for this work is publicly available.
Authors: Christiaan Lamers, Ahmed Nabil Belbachir, Thomas B\"ack, Niki van Stein
Abstract: Catastrophic forgetting can be trivially alleviated by keeping all data from previous tasks in memory. Therefore, minimizing the memory footprint while maximizing the amount of relevant information is crucial to the challenge of continual learning. This paper aims to decrease required memory for memory-based continuous learning algorithms. We explore the options of extracting a minimal amount of information, while maximally alleviating forgetting. We propose the usage of lightweight generators based on Singular Value Decomposition to enhance existing continual learning methods, such as A-GEM and Experience Replay. These generators need a minimal amount of memory while being maximally effective. They require no training time, just a single linear-time fitting step, and can capture a distribution effectively from a small number of data samples. Depending on the dataset and network architecture, our results show a significant increase in average accuracy compared to the original methods. Our method shows great potential in minimizing the memory footprint of memory-based continual learning algorithms.
Authors: Alberto Pirillo, Luca Colombo
Abstract: Growing concerns over data privacy underscore the need for deep learning methods capable of processing sensitive information without compromising confidentiality. Among privacy-enhancing technologies, Homomorphic Encryption (HE) stands out by providing post-quantum cryptographic security and end-to-end data protection, safeguarding data even during computation. While Deep Neural Networks (DNNs) have gained attention in HE settings, their use has largely been restricted to encrypted inference. Prior research on encrypted training has primarily focused on logistic regression or has relied on multi-party computation to enable model fine-tuning. This stems from the substantial computational overhead and algorithmic complexity involved in DNNs training under HE. In this paper, we present ReBoot, the first framework to enable fully encrypted and non-interactive training of DNNs. Built upon the CKKS scheme, ReBoot introduces a novel HE-compliant neural network architecture based on local error signals, specifically designed to minimize multiplicative depth and reduce noise accumulation. ReBoot employs a tailored packing strategy that leverages real-number arithmetic via SIMD operations, significantly lowering both computational and memory overhead. Furthermore, by integrating approximate bootstrapping, ReBoot learning algorithm supports effective training of arbitrarily deep multi-layer perceptrons, making it well-suited for machine learning as-a-service. ReBoot is evaluated on both image recognition and tabular benchmarks, achieving accuracy comparable to 32-bit floating-point plaintext training while enabling fully encrypted training. It improves test accuracy by up to +3.27% over encrypted logistic regression, and up to +6.83% over existing encrypted DNN frameworks, while reducing training latency by up to 8.83x. ReBoot is made available to the scientific community as a public repository.
Authors: Jungwoo Park, Taewhoo Lee, Chanwoong Yoon, Hyeon Hwang, Jaewoo Kang
Abstract: Extreme activation outliers in Large Language Models (LLMs) critically degrade quantization performance, hindering efficient on-device deployment. While channel-wise operations and adaptive gradient scaling are recognized causes, practical mitigation remains challenging. We introduce Outlier-Safe Pre-Training (OSP), a practical guideline that proactively prevents outlier formation rather than relying on post-hoc mitigation. OSP combines three key innovations: (1) the Muon optimizer, eliminating privileged bases while maintaining training efficiency; (2) Single-Scale RMSNorm, preventing channel-wise amplification; and (3) a learnable embedding projection, redistributing activation magnitudes originating from embedding matrices. We validate OSP by training a 1.4B-parameter model on 1 trillion tokens, which is the first production-scale LLM trained without such outliers. Under aggressive 4-bit quantization, our OSP model achieves a 35.7 average score across 10 benchmarks (compared to 26.5 for an Adam-trained model), with only a 2% training overhead. Remarkably, OSP models exhibit near-zero excess kurtosis (0.04) compared to extreme values (1818.56) in standard models, fundamentally altering LLM quantization behavior. Our work demonstrates that outliers are not inherent to LLMs but are consequences of training strategies, paving the way for more efficient LLM deployment. The source code and pretrained checkpoints are available at https://github.com/dmis-lab/Outlier-Safe-Pre-Training.
URLs: https://github.com/dmis-lab/Outlier-Safe-Pre-Training.
Authors: Nathan Maurer, Harshal Kaushik, Roshni Anna Jacob, Jie Zhang, Souma Chowdhury
Abstract: The resilience of critical infrastructure networks (CINs) after disruptions, such as those caused by natural hazards, depends on both the speed of restoration and the extent to which operational functionality can be regained. Allocating resources for restoration is a combinatorial optimal planning problem that involves determining which crews will repair specific network nodes and in what order. This paper presents a novel graph-based formulation that merges two interconnected graphs, representing crew and transportation nodes and power grid nodes, into a single heterogeneous graph. To enable efficient planning, graph reinforcement learning (GRL) is integrated with bigraph matching. GRL is utilized to design the incentive function for assigning crews to repair tasks based on the graph-abstracted state of the environment, ensuring generalization across damage scenarios. Two learning techniques are employed: a graph neural network trained using Proximal Policy Optimization and another trained via Neuroevolution. The learned incentive functions inform a bipartite graph that links crews to repair tasks, enabling weighted maximum matching for crew-to-task allocations. An efficient simulation environment that pre-computes optimal node-to-node path plans is used to train the proposed restoration planning methods. An IEEE 8500-bus power distribution test network coupled with a 21 square km transportation network is used as the case study, with scenarios varying in terms of numbers of damaged nodes, depots, and crews. Results demonstrate the approach's generalizability and scalability across scenarios, with learned policies providing 3-fold better performance than random policies, while also outperforming optimization-based solutions in both computation time (by several orders of magnitude) and power restored.
Authors: Seyedmorteza Sadat, Tobias Vontobel, Farnood Salehi, Romann M. Weber
Abstract: Classifier-free guidance (CFG) has become an essential component of modern conditional diffusion models. Although highly effective in practice, the underlying mechanisms by which CFG enhances quality, detail, and prompt alignment are not fully understood. We present a novel perspective on CFG by analyzing its effects in the frequency domain, showing that low and high frequencies have distinct impacts on generation quality. Specifically, low-frequency guidance governs global structure and condition alignment, while high-frequency guidance mainly enhances visual fidelity. However, applying a uniform scale across all frequencies -- as is done in standard CFG -- leads to oversaturation and reduced diversity at high scales and degraded visual quality at low scales. Based on these insights, we propose frequency-decoupled guidance (FDG), an effective approach that decomposes CFG into low- and high-frequency components and applies separate guidance strengths to each component. FDG improves image quality at low guidance scales and avoids the drawbacks of high CFG scales by design. Through extensive experiments across multiple datasets and models, we demonstrate that FDG consistently enhances sample fidelity while preserving diversity, leading to improved FID and recall compared to CFG, establishing our method as a plug-and-play alternative to standard classifier-free guidance.
Authors: Carlos Stein Brito
Abstract: Deep neural networks require principled uncertainty quantification, yet existing variational inference methods often employ isotropic Gaussian approximations in weight space that poorly match the network's inherent geometry. We address this mismatch by introducing Concentration-Adapted Perturbations (CAP), a variational framework that models weight uncertainties directly on the unit hypersphere using von Mises-Fisher distributions. Building on recent work in radial-directional posterior decompositions and spherical weight constraints, CAP provides the first complete theoretical framework connecting directional statistics to practical noise regularization in neural networks. Our key contribution is an analytical derivation linking vMF concentration parameters to activation noise variance, enabling each layer to learn its optimal uncertainty level through a novel closed-form KL divergence regularizer. In experiments on CIFAR-10, CAP significantly improves model calibration - reducing Expected Calibration Error by 5.6x - while providing interpretable layer-wise uncertainty profiles. CAP requires minimal computational overhead and integrates seamlessly into standard architectures, offering a theoretically grounded yet practical approach to uncertainty quantification in deep learning.
Authors: Shrey Dixit, Kayson Fakhar, Fatemeh Hadaeghi, Patrick Mineault, Konrad P. Kording, Claus C. Hilgetag
Abstract: Neural networks now generate text, images, and speech with billions of parameters, producing a need to know how each neural unit contributes to these high-dimensional outputs. Existing explainable-AI methods, such as SHAP, attribute importance to inputs, but cannot quantify the contributions of neural units across thousands of output pixels, tokens, or logits. Here we close that gap with Multiperturbation Shapley-value Analysis (MSA), a model-agnostic game-theoretic framework. By systematically lesioning combinations of units, MSA yields Shapley Modes, unit-wise contribution maps that share the exact dimensionality of the model's output. We apply MSA across scales, from multi-layer perceptrons to the 56-billion-parameter Mixtral-8x7B and Generative Adversarial Networks (GAN). The approach demonstrates how regularisation concentrates computation in a few hubs, exposes language-specific experts inside the LLM, and reveals an inverted pixel-generation hierarchy in GANs. Together, these results showcase MSA as a powerful approach for interpreting, editing, and compressing deep neural networks.
Authors: Ben Keslaki
Abstract: Modern deep learning architectures excel at optimization, but only after the data has entered the network. The true bottleneck lies in preparing the right input: minimal, salient, and structured in a way that reflects the essential patterns of the data. We propose DRIFT (Data Reduction via Informative Feature Transformation), a novel preprocessing technique inspired by vibrational analysis in physical systems, to identify and extract the most resonant modes of input data prior to training. Unlike traditional models that attempt to learn amidst both signal and noise, DRIFT mimics physics perception by emphasizing informative features while discarding irrelevant elements. The result is a more compact and interpretable representation that enhances training stability and generalization performance. In DRIFT, images are projected onto a low-dimensional basis formed by spatial vibration mode shapes of plates, offering a physically grounded feature set. This enables neural networks to operate with drastically fewer input dimensions (~ 50 features on MNIST and less than 100 on CIFAR100) while achieving competitive classification accuracy. Extensive experiments across MNIST and CIFAR100 demonstrate DRIFT's superiority over standard pixel-based models and PCA in terms of training stability, resistance to overfitting, and generalization robustness. Notably, DRIFT displays minimal sensitivity to changes in batch size, network architecture, and image resolution, further establishing it as a resilient and efficient data representation strategy. This work shifts the focus from architecture engineering to input curation and underscores the power of physics-driven data transformations in advancing deep learning performance.
Authors: Yihong Luo, Shuchen Xue, Tianyang Hu, Jing Tang
Abstract: The pursuit of efficient and controllable high-quality content generation remains a central challenge in artificial intelligence-generated content (AIGC). While one-step generators, enabled by diffusion distillation techniques, offer excellent generation quality and computational efficiency, adapting them to new control conditions--such as structural constraints, semantic guidelines, or external inputs--poses a significant challenge. Conventional approaches often necessitate computationally expensive modifications to the base model and subsequent diffusion distillation. This paper introduces Noise Consistency Training (NCT), a novel and lightweight approach to directly integrate new control signals into pre-trained one-step generators without requiring access to original training images or retraining the base diffusion model. NCT operates by introducing an adapter module and employs a noise consistency loss in the noise space of the generator. This loss aligns the adapted model's generation behavior across noises that are conditionally dependent to varying degrees, implicitly guiding it to adhere to the new control. Theoretically, this training objective can be understood as minimizing the distributional distance between the adapted generator and the conditional distribution induced by the new conditions. NCT is modular, data-efficient, and easily deployable, relying only on the pre-trained one-step generator and a control signal model. Extensive experiments demonstrate that NCT achieves state-of-the-art controllable generation in a single forward pass, surpassing existing multi-step and distillation-based methods in both generation quality and computational efficiency. Code is available at https://github.com/Luo-Yihong/NCT
Authors: Emmeran Johnson, David Mart\'inez-Rubio, Ciara Pike-Burke, Patrick Rebeschini
Abstract: We study online convex optimization on $\ell_p$-balls in $\mathbb{R}^d$ for $p > 2$. While always sub-linear, the optimal regret exhibits a shift between the high-dimensional setting ($d > T$), when the dimension $d$ is greater than the time horizon $T$ and the low-dimensional setting ($d \leq T$). We show that Follow-the-Regularised-Leader (FTRL) with time-varying regularisation which is adaptive to the dimension regime is anytime optimal for all dimension regimes. Motivated by this, we ask whether it is possible to obtain anytime optimality of FTRL with fixed non-adaptive regularisation. Our main result establishes that for separable regularisers, adaptivity in the regulariser is necessary, and that any fixed regulariser will be sub-optimal in one of the two dimension regimes. Finally, we provide lower bounds which rule out sub-linear regret bounds for the linear bandit problem in sufficiently high-dimension for all $\ell_p$-balls with $p \geq 1$.
Authors: Carlos Stein Brito
Abstract: Model regularization requires extensive manual tuning to balance complexity against overfitting. Cross-regularization resolves this tradeoff by directly adapting regularization parameters through validation gradients during training. The method splits parameter optimization - training data guides feature learning while validation data shapes complexity controls - converging provably to cross-validation optima. When implemented through noise injection in neural networks, this approach reveals striking patterns: unexpectedly high noise tolerance and architecture-specific regularization that emerges organically during training. Beyond complexity control, the framework integrates seamlessly with data augmentation, uncertainty calibration and growing datasets while maintaining single-run efficiency through a simple gradient-based approach.
Authors: Yuhui Sun (University of Alberta), Xiyao Wang (University of Toronto), Zixi Li (Zhejiang University), Jinman Zhao (University of Toronto)
Abstract: While large-scale unsupervised language models (LMs) capture broad world knowledge and reasoning capabilities, steering their behavior toward desired objectives remains challenging due to the lack of explicit supervision. Existing alignment techniques, such as reinforcement learning from human feedback (RLHF), rely on training a reward model and performing reinforcement learning to align with human preferences. However, RLHF is often computationally intensive, unstable, and sensitive to hyperparameters. To address these limitations, Direct Preference Optimization (DPO) was introduced as a lightweight and stable alternative, enabling direct alignment of language models with pairwise preference data via classification loss. However, DPO and its extensions generally assume a single static preference distribution, limiting flexibility in multi-objective or dynamic alignment settings. In this paper, we propose a novel framework: Multi-Preference Lambda-weighted Listwise DPO, which extends DPO to incorporate multiple human preference dimensions (e.g., helpfulness, harmlessness, informativeness) and enables dynamic interpolation through a controllable simplex-weighted formulation. Our method supports both listwise preference feedback and flexible alignment across varying user intents without re-training. Empirical and theoretical analysis demonstrates that our method is as effective as traditional DPO on static objectives while offering greater generality and adaptability for real-world deployment.
Authors: Chenhao Si, Ming Yan
Abstract: Physics-informed neural networks (PINNs) are extensively employed to solve partial differential equations (PDEs) by ensuring that the outputs and gradients of deep learning models adhere to the governing equations. However, constrained by computational limitations, PINNs are typically optimized using a finite set of points, which poses significant challenges in guaranteeing their convergence and accuracy. In this study, we proposed a new weighting scheme that will adaptively change the weights to the loss functions from isolated points to their continuous neighborhood regions. The empirical results show that our weighting scheme can reduce the relative $L^2$ errors to a lower value.
Authors: Vanessa Kosoy
Abstract: We propose a new variant of online learning that we call "ambiguous online learning". In this setting, the learner is allowed to produce multiple predicted labels. Such an "ambiguous prediction" is considered correct when at least one of the labels is correct, and none of the labels are "predictably wrong". The definition of "predictably wrong" comes from a hypothesis class in which hypotheses are also multi-valued. Thus, a prediction is "predictably wrong" if it's not allowed by the (unknown) true hypothesis. In particular, this setting is natural in the context of multivalued dynamical systems, recommendation algorithms and lossless compression. It is also strongly related to so-called "apple tasting". We show that in this setting, there is a trichotomy of mistake bounds: up to logarithmic factors, any hypothesis class has an optimal mistake bound of either Theta(1), Theta(sqrt(N)) or N.
Authors: Eurico Covas
Abstract: Art curatorship has always been mostly the subjective work of human experts, who, with extensive knowledge of many and diverse artworks, select a few of those to present in communal spaces, spaces that evolved into what we now call art galleries. There are no hard and fast set of rules on how to select these artworks, given a theme which either is presented to the art curator or constructed by her/him. Here we present a series of artificial models -- a total of four related models -- based on machine learning techniques (a subset of artificial intelligence) that attempt to learn from existing exhibitions which have been curated by human experts, in order to be able to do similar curatorship work. We focus exclusively on the last 25 years of past exhibitions at the Metropolitan Museum of Art in New York, due to the quality of the data available and the physical and time limitations of our research. Our four artificial intelligence models achieve a reasonable ability at imitating these various curators responsible for all those exhibitions, with various degrees of precision and curatorial coherence. In particular, we can conclude two key insights: first, that there is sufficient information in these exhibitions to construct an artificial intelligence model that replicates past exhibitions with an accuracy well above random choices; second, that using feature engineering and carefully designing the architecture of modest size models can make them as good as those using the so-called large language models such as GPT in a brute force approach. We also believe, based on small attempts to use the models in out-of-sample experiments, that given more much more data, it should be possible for these kinds of artificial intelligence agents to be closer and closer to the aesthetic and curatorial judgment of human art curators.
Authors: Miles Wang, Tom Dupr\'e la Tour, Olivia Watkins, Alex Makelov, Ryan A. Chi, Samuel Miserendino, Johannes Heidecke, Tejal Patwardhan, Dan Mossing
Abstract: Understanding how language models generalize behaviors from their training to a broader deployment distribution is an important problem in AI safety. Betley et al. discovered that fine-tuning GPT-4o on intentionally insecure code causes "emergent misalignment," where models give stereotypically malicious responses to unrelated prompts. We extend this work, demonstrating emergent misalignment across diverse conditions, including reinforcement learning on reasoning models, fine-tuning on various synthetic datasets, and in models without safety training. To investigate the mechanisms behind this generalized misalignment, we apply a "model diffing" approach using sparse autoencoders to compare internal model representations before and after fine-tuning. This approach reveals several "misaligned persona" features in activation space, including a toxic persona feature which most strongly controls emergent misalignment and can be used to predict whether a model will exhibit such behavior. Additionally, we investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.
Authors: Yichao Fu, Rui Ge, Zelei Shao, Zhijie Deng, Hao Zhang
Abstract: Reasoning models excel by generating long chain-of-thoughts, but decoding the resulting thousands of tokens is slow. Token-level speculative decoding (SD) helps, but its benefit is capped, because the chance that an entire $\gamma$-token guess is correct falls exponentially as $\gamma$ grows. This means allocating more compute for longer token drafts faces an algorithmic ceiling -- making the speedup modest and hardware-agnostic. We raise this ceiling with Lookahead Reasoning, which exploits a second, step-level layer of parallelism. Our key insight is that reasoning models generate step-by-step, and each step needs only to be semantically correct, not exact token matching. In Lookahead Reasoning, a lightweight draft model proposes several future steps; the target model expands each proposal in one batched pass, and a verifier keeps semantically correct steps while letting the target regenerate any that fail. Token-level SD still operates within each reasoning step, so the two layers of parallelism multiply. We show Lookahead Reasoning lifts the peak speedup of SD both theoretically and empirically. Across GSM8K, AIME, and other benchmarks, Lookahead Reasoning improves the speedup of SD from 1.4x to 2.1x while preserving answer quality, and its speedup scales better with additional GPU throughput. Our code is available at https://github.com/hao-ai-lab/LookaheadReasoning
Authors: Zeju Qiu, Weiyang Liu, Adrian Weller, Bernhard Sch\"olkopf
Abstract: Orthogonal finetuning (OFT) offers highly parameter-efficient adaptation while preventing catastrophic forgetting, but its high runtime and memory demands limit practical deployment. We identify the core computational bottleneck in OFT as its weight-centric implementation, which relies on costly matrix-matrix multiplications with cubic complexity. To overcome this, we propose OFTv2, an input-centric reformulation that instead uses matrix-vector multiplications (i.e., matrix-free computation), reducing the computational cost to quadratic. We further introduce the Cayley-Neumann parameterization, an efficient orthogonal parameterization that approximates the matrix inversion in Cayley transform via a truncated Neumann series. These modifications allow OFTv2 to achieve up to 10x faster training and 3x lower GPU memory usage without compromising performance. In addition, we extend OFTv2 to support finetuning quantized foundation models and show that it outperforms the popular QLoRA in training stability, efficiency, and memory usage.
Authors: Pablo Moriano, Steven C. Hespeler, Mingyan Li, Maria Mahbub
Abstract: Modern cyberattacks in cyber-physical systems (CPS) rapidly evolve and cannot be deterred effectively with most current methods which focused on characterizing past threats. Adaptive anomaly detection (AAD) is among the most promising techniques to detect evolving cyberattacks focused on fast data processing and model adaptation. AAD has been researched in the literature extensively; however, to the best of our knowledge, our work is the first systematic literature review (SLR) on the current research within this field. We present a comprehensive SLR, gathering 397 relevant papers and systematically analyzing 65 of them (47 research and 18 survey papers) on AAD in CPS studies from 2013 to 2023 (November). We introduce a novel taxonomy considering attack types, CPS application, learning paradigm, data management, and algorithms. Our analysis indicates, among other findings, that reviewed works focused on a single aspect of adaptation (either data processing or model adaptation) but rarely in both at the same time. We aim to help researchers to advance the state of the art and help practitioners to become familiar with recent progress in this field. We identify the limitations of the state of the art and provide recommendations for future research directions.
Authors: Jason Benjamin Baretz, Max Fieg, Vijay Ganesh, Aishik Ghosh, V. Knapp-Perez, Jake Rudolph, Daniel Whiteson
Abstract: Particle physics theories, such as those which explain neutrino flavor mixing, arise from a vast landscape of model-building possibilities. A model's construction typically relies on the intuition of theorists. It also requires considerable effort to identify appropriate symmetry groups, assign field representations, and extract predictions for comparison with experimental data. We develop an Autonomous Model Builder (AMBer), a framework in which a reinforcement learning agent interacts with a streamlined physics software pipeline to search these spaces efficiently. AMBer selects symmetry groups, particle content, and group representation assignments to construct viable models while minimizing the number of free parameters introduced. We validate our approach in well-studied regions of theory space and extend the exploration to a novel, previously unexamined symmetry group. While demonstrated in the context of neutrino flavor theories, this approach of reinforcement learning with physics software feedback may be extended to other theoretical model-building problems in the future.
Authors: Siyang Song, Yupeng Huo, Shiqing Tang, Jiaee Cheong, Rui Gao, Michel Valstar, Hatice Gunes
Abstract: Depression is a common mental illness across current human society. Traditional depression assessment relying on inventories and interviews with psychologists frequently suffer from subjective diagnosis results, slow and expensive diagnosis process as well as lack of human resources. Since there is a solid evidence that depression is reflected by various human internal brain activities and external expressive behaviours, early traditional machine learning (ML) and advanced deep learning (DL) models have been widely explored for human behaviour-based automatic depression assessment (ADA) since 2012. However, recent ADA surveys typically only focus on a limited number of human behaviour modalities. Despite being used as a theoretical basis for developing ADA approaches, existing ADA surveys lack a comprehensive review and summary of multi-modal depression-related human behaviours. To bridge this gap, this paper specifically summarises depression-related human behaviours across a range of modalities (e.g. the human brain, verbal language and non-verbal audio/facial/body behaviours). We focus on conducting an up-to-date and comprehensive survey of ML-based ADA approaches for learning depression cues from these behaviours as well as discussing and comparing their distinctive features and limitations. In addition, we also review existing ADA competitions and datasets, identify and discuss the main challenges and opportunities to provide further research directions for future ADA researchers.
Authors: Michael Williams
Abstract: In this work, we investigate how autonomous agents, organized into tribes, learn to use communication signals to coordinate their activities and enhance their collective efficiency. Using the NEC-DAC (Neurally Encoded Culture - Distributed Autonomous Communicators) system, where each agent is equipped with its own neural network for decision-making, we demonstrate how these agents develop a shared behavioral system -- akin to a culture -- through learning and signalling. Our research focuses on the self-organization of culture within these tribes of agents and how varying communication strategies impact their fitness and cooperation. By analyzing different social structures, such as authority hierarchies, we show that the culture of cooperation significantly influences the tribe's performance. Furthermore, we explore how signals not only facilitate the emergence of culture but also enable its transmission across generations of agents. Additionally, we examine the benefits of coordinating behavior and signaling within individual agents' neural networks.
Authors: Chong Di, Shuwang Zhou, Da Chen, Jean-Marie Mirebeau, Minglei Shu, Laurent D. Cohen
Abstract: The computation of minimal paths for the applications in tracking tubular structures such as blood vessels and roads is challenged by complex morphologies and environmental variations. Existing approaches can be roughly categorized into two research lines: the point-wise based models and the segment-wise based models. Although segment-wise approaches have obtained promising results in many scenarios, they often suffer from computational inefficiency and heavily rely on a prescribed prior to fit the target elongated shapes. We propose a novel framework that casts segment-wise tracking as a Markov Decision Process (MDP), enabling a reinforcement learning approach. Our method leverages Q-Learning to dynamically explore a graph of segments, computing edge weights on-demand and adaptively expanding the search space. This strategy avoids the high cost of a pre-computed graph and proves robust to incomplete initial information. Experimental reuslts on typical tubular structure datasets demonstrate that our method significantly outperforms state-of-the-art point-wise and segment-wise approaches. The proposed method effectively handles complex topologies and maintains global path coherence without depending on extensive prior structural knowledge.
Authors: Diego Di Carlo (RIKEN AIP), Mathieu Fontaine (LTCI, IP Paris), Aditya Arie Nugraha (RIKEN AIP), Yoshiaki Bando (RIKEN AIP), Kazuyoshi Yoshii
Abstract: This paper describes a sound source localization (SSL) technique that combines an $\alpha$-stable model for the observed signal with a neural network-based approach for modeling steering vectors. Specifically, a physics-informed neural network, referred to as Neural Steerer, is used to interpolate measured steering vectors (SVs) on a fixed microphone array. This allows for a more robust estimation of the so-called $\alpha$-stable spatial measure, which represents the most plausible direction of arrival (DOA) of a target signal. As an $\alpha$-stable model for the non-Gaussian case ($\alpha$ $\in$ (0, 2)) theoretically defines a unique spatial measure, we choose to leverage it to account for residual reconstruction error of the Neural Steerer in the downstream tasks. The objective scores indicate that our proposed technique outperforms state-of-the-art methods in the case of multiple sound sources.
Authors: Sheraz Khan, Subha Madhavan, Kannan Natarajan
Abstract: The recent work by Shojaee et al. (2025), titled The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity, presents a compelling empirical finding, a reasoning cliff, where the performance of Large Reasoning Models (LRMs) collapses beyond a specific complexity threshold, which the authors posit as an intrinsic scaling limitation of Chain-of-Thought (CoT) reasoning. This commentary, while acknowledging the study's methodological rigor, contends that this conclusion is confounded by experimental artifacts. We argue that the observed failure is not evidence of a fundamental cognitive boundary, but rather a predictable outcome of system-level constraints in the static, text-only evaluation paradigm, including tool use restrictions, context window recall issues, the absence of crucial cognitive baselines, inadequate statistical reporting, and output generation limits. We reframe this performance collapse through the lens of an agentic gap, asserting that the models are not failing at reasoning, but at execution within a profoundly restrictive interface. We empirically substantiate this critique by demonstrating a striking reversal. A model, initially declaring a puzzle impossible when confined to text-only generation, now employs agentic tools to not only solve it but also master variations of complexity far beyond the reasoning cliff it previously failed to surmount. Additionally, our empirical analysis of tool-enabled models like o4-mini and GPT-4o reveals a hierarchy of agentic reasoning, from simple procedural execution to complex meta-cognitive self-correction, which has significant implications for how we define and measure machine intelligence. The illusion of thinking attributed to LRMs is less a reasoning deficit and more a consequence of an otherwise capable mind lacking the tools for action.
Authors: Weizhi Zhang, Yangning Li, Yuanchen Bei, Junyu Luo, Guancheng Wan, Liangwei Yang, Chenxuan Xie, Yuyao Yang, Wei-Chieh Huang, Chunyu Miao, Henry Peng Zou, Xiao Luo, Yusheng Zhao, Yankai Chen, Chunkit Chan, Peilin Zhou, Xinyang Zhang, Chenwei Zhang, Jingbo Shang, Ming Zhang, Yangqiu Song, Irwin King, Philip S. Yu
Abstract: Information retrieval is a cornerstone of modern knowledge acquisition, enabling billions of queries each day across diverse domains. However, traditional keyword-based search engines are increasingly inadequate for handling complex, multi-step information needs. Our position is that Large Language Models (LLMs), endowed with reasoning and agentic capabilities, are ushering in a new paradigm termed Agentic Deep Research. These systems transcend conventional information search techniques by tightly integrating autonomous reasoning, iterative retrieval, and information synthesis into a dynamic feedback loop. We trace the evolution from static web search to interactive, agent-based systems that plan, explore, and learn. We also introduce a test-time scaling law to formalize the impact of computational depth on reasoning and search. Supported by benchmark results and the rise of open-source implementations, we demonstrate that Agentic Deep Research not only significantly outperforms existing approaches, but is also poised to become the dominant paradigm for future information seeking. All the related resources, including industry products, research papers, benchmark datasets, and open-source implementations, are collected for the community in https://github.com/DavidZWZ/Awesome-Deep-Research.
Authors: Soojin Park, Suyeon Kang, Chioun Lee
Abstract: Causal decomposition analysis aims to assess the effect of modifying risk factors on reducing social disparities in outcomes. Recently, this analysis has incorporated individual characteristics when modifying risk factors by utilizing optimal treatment regimes (OTRs). Since the newly defined individualized effects rely on the no omitted confounding assumption, developing sensitivity analyses to account for potential omitted confounding is essential. Moreover, OTRs and individualized effects are primarily based on binary risk factors, and no formal approach currently exists to benchmark the strength of omitted confounding using observed covariates for binary risk factors. To address this gap, we extend a simulation-based sensitivity analysis that simulates unmeasured confounders, addressing two sources of bias emerging from deriving OTRs and estimating individualized effects. Additionally, we propose a formal bounding strategy that benchmarks the strength of omitted confounding for binary risk factors. Using the High School Longitudinal Study 2009 (HSLS:09), we demonstrate this sensitivity analysis and benchmarking method.
Authors: Sivaraman Balakrishnan, Tudor Manole, Larry Wasserman
Abstract: In many applications of optimal transport (OT), the object of primary interest is the optimal transport map. This map rearranges mass from one probability distribution to another in the most efficient way possible by minimizing a specified cost. In this paper we review recent advances in estimating and developing limit theorems for the OT map, using samples from the underlying distributions. We also review parallel lines of work that establish similar results for special cases and variants of the basic OT setup. We conclude with a discussion of key directions for future research with the goal of providing practitioners with reliable inferential tools.
Authors: Chen Zeno, Hila Manor, Greg Ongie, Nir Weinberger, Tomer Michaeli, Daniel Soudry
Abstract: While diffusion models generate high-quality images via probability flow, the theoretical understanding of this process remains incomplete. A key question is when probability flow converges to training samples or more general points on the data manifold. We analyze this by studying the probability flow of shallow ReLU neural network denoisers trained with minimal $\ell^2$ norm. For intuition, we introduce a simpler score flow and show that for orthogonal datasets, both flows follow similar trajectories, converging to a training point or a sum of training points. However, early stopping by the diffusion time scheduler allows probability flow to reach more general manifold points. This reflects the tendency of diffusion models to both memorize training samples and generate novel points that combine aspects of multiple samples, motivating our study of such behavior in simplified settings. We extend these results to obtuse simplex data and, through simulations in the orthogonal case, confirm that probability flow converges to a training point, a sum of training points, or a manifold point. Moreover, memorization decreases when the number of training samples grows, as fewer samples accumulate near training points.
Authors: Omer Luxembourg, Haim Permuter, Eliya Nachmani
Abstract: Masked diffusion language models (MDLM) have shown strong promise for non-autoregressive text generation, yet existing samplers act as implicit planners, selecting tokens to unmask via denoiser confidence or entropy scores. Such heuristics falter under parallel unmasking - they ignore pairwise interactions between tokens and cannot account for dependencies when unmasking multiple positions at once, limiting their inference time to traditional auto-regressive (AR) models. We introduce the Dilated-scheduled Unmasking Strategy (DUS), an inference-only, planner-model-free method that requires no additional training. DUS leverages a first-order Markov assumption to partition sequence positions into dilation-based groups of non-adjacent tokens, enabling independent, parallel unmasking steps that respect local context that minimizes the joint entropy of each iteration step. Unlike semi-AR block approaches (e.g., LLADA and Dream) that still invoke the denoiser per block, DUS reduces the number of denoiser calls to O(log B) per generation block - yielding substantial speedup over the O(B) run time of state-of-the-art diffusion models, where B is the block size in the semi-AR inference process. In experiments on math (GSM8K) and code completion (Humaneval, MBPP) benchmarks - domains suited to non-ordinal generation - DUS improves scores over parallel confidence-based planner, without modifying the underlying denoiser. DUS offers a lightweight, budget-aware approach to efficient, high-quality text generation, paving the way to unlock the true capabilities of MDLMs.
Authors: Vincent Leon, S. Rasoul Etesami
Abstract: We consider the problem of online dynamic mechanism design for sequential auctions in unknown environments, where the underlying market and, thus, the bidders' values vary over time as interactions between the seller and the bidders progress. We model the sequential auctions as an infinite-horizon average-reward Markov decision process (MDP), where the transition kernel and reward functions are unknown to the seller. In each round, the seller determines an allocation and a payment for each bidder. Each bidder receives a private reward and submits a sealed bid to the seller. The state, which represents the underlying market, evolves according to an unknown transition kernel and the seller's allocation policy. Unlike existing works that formulate the problem as a multi-armed bandit model or as an episodic MDP, where the environment resets to an initial state after each round or episode, our paper considers a more realistic and sophisticated setting in which the market continues to evolve without restarting. We first extend the Vickrey-Clarke-Groves (VCG) mechanism, which is known to be efficient, truthful, and individually rational for one-shot static auctions, to sequential auctions, thereby obtaining a dynamic VCG mechanism counterpart that preserves these desired properties. We then focus on the online setting and develop an online reinforcement learning algorithm for the seller to learn the underlying MDP model and implement a mechanism that closely resembles the dynamic VCG mechanism. We show that the learned online mechanism asymptotically converges to a dynamic mechanism that approximately satisfies efficiency, truthfulness, and individual rationality with arbitrarily high probability and achieves guaranteed performance in terms of various notions of regret.
Authors: Xinlin Wang, Mats Brorsson
Abstract: Uplift modeling has achieved significant success in various fields, particularly in online marketing. It is a method that primarily utilizes machine learning and deep learning to estimate individual treatment effects. This paper we apply uplift modeling to analyze the effect of company adjustment on their financial status, and we treat these adjustment as treatments or interventions in this study. Although there have been extensive studies and application regarding binary treatments, multiple treatments, and continuous treatments, company adjustment are often more complex than these scenarios, as they constitute a series of multiple time-dependent actions. The effect estimation of company adjustment needs to take into account not only individual treatment traits but also the temporal order of this series of treatments. This study collects a real-world data set about company financial statements and reported behavior in Luxembourg for the experiments. First, we use two meta-learners and three other well-known uplift models to analyze different company adjustment by simplifying the adjustment as binary treatments. Furthermore, we propose a new uplift modeling framework (MTDnet) to address the time-dependent nature of these adjustment, and the experimental result shows the necessity of considering the timing of these adjustment.
Authors: Dan Garber
Abstract: In was recently established that for convex optimization problems with a sparse optimal solution (may it be entry-wise sparsity or matrix rank-wise sparsity) it is possible to have linear convergence rates which depend on an improved mixed-norm condition number of the form $\frac{\beta_1{}s}{\alpha_2}$, where $\beta_1$ is the $\ell_1$-Lipchitz continuity constant of the gradient, $\alpha_2$ is the $\ell_2$-quadratic growth constant, and $s$ is the sparsity of the optimal solution. However, beyond the improved convergence rate, these methods are unable to leverage the sparsity of optimal solutions towards improving also the runtime of each iteration, which may still be prohibitively high for high-dimensional problems. In this work, we establish that linear convergence rates which depend on this improved condition number can be obtained using only sparse updates, which may result in overall significantly improved running times. Moreover, our methods are considerably easier to implement.
Authors: Christopher Agia, Rohan Sinha, Jingyun Yang, Rika Antonova, Marco Pavone, Haruki Nishimura, Masha Itkina, Jeannette Bohg
Abstract: In robot imitation learning, policy performance is tightly coupled with the quality and composition of the demonstration data. Yet, developing a precise understanding of how individual demonstrations contribute to downstream outcomes - such as closed-loop task success or failure - remains a persistent challenge. We propose CUPID, a robot data curation method based on a novel influence function-theoretic formulation for imitation learning policies. Given a set of evaluation rollouts, CUPID estimates the influence of each training demonstration on the policy's expected return. This enables ranking and selection of demonstrations according to their impact on the policy's closed-loop performance. We use CUPID to curate data by 1) filtering out training demonstrations that harm policy performance and 2) subselecting newly collected trajectories that will most improve the policy. Extensive simulated and hardware experiments show that our approach consistently identifies which data drives test-time performance. For example, training with less than 33% of curated data can yield state-of-the-art diffusion policies on the simulated RoboMimic benchmark, with similar gains observed in hardware. Furthermore, hardware experiments show that our method can identify robust strategies under distribution shift, isolate spurious correlations, and even enhance the post-training of generalist robot policies. Additional materials are made available at: https://cupid-curation.github.io.
Authors: Bruno Aristimunha, Dung Truong, Pierre Guetschel, Seyed Yahya Shirazi, Isabelle Guyon, Alexandre R. Franco, Michael P. Milham, Aviv Dotan, Scott Makeig, Alexandre Gramfort, Jean-Remi King, Marie-Constance Corsi, Pedro A. Vald\'es-Sosa, Amit Majumdar, Alan Evans, Terrence J Sejnowski, Oren Shriki, Sylvain Chevallier, Arnaud Delorme
Abstract: Current electroencephalogram (EEG) decoding models are typically trained on small numbers of subjects performing a single task. Here, we introduce a large-scale, code-submission-based competition comprising two challenges. First, the Transfer Challenge asks participants to build and test a model that can zero-shot decode new tasks and new subjects from their EEG data. Second, the Psychopathology factor prediction Challenge asks participants to infer subject measures of mental health from EEG data. For this, we use an unprecedented, multi-terabyte dataset of high-density EEG signals (128 channels) recorded from over 3,000 child to young adult subjects engaged in multiple active and passive tasks. We provide several tunable neural network baselines for each of these two challenges, including a simple network and demographic-based regression models. Developing models that generalise across tasks and individuals will pave the way for ML network architectures capable of adapting to EEG data collected from diverse tasks and individuals. Similarly, predicting mental health-relevant personality trait values from EEG might identify objective biomarkers useful for clinical diagnosis and design of personalised treatment for psychological conditions. Ultimately, the advances spurred by this challenge could contribute to the development of computational psychiatry and useful neurotechnology, and contribute to breakthroughs in both fundamental neuroscience and applied clinical research.
Authors: Kyeongwon Lee, Lizhen Lin, Jaewoo Park, Seonghyun Jeong
Abstract: This work establishes that sparse Bayesian neural networks achieve optimal posterior contraction rates over anisotropic Besov spaces and their hierarchical compositions. These structures reflect the intrinsic dimensionality of the underlying function, thereby mitigating the curse of dimensionality. Our analysis shows that Bayesian neural networks equipped with either sparse or continuous shrinkage priors attain the optimal rates which are dependent on the intrinsic dimension of the true structures. Moreover, we show that these priors enable rate adaptation, allowing the posterior to contract at the optimal rate even when the smoothness level of the true function is unknown. The proposed framework accommodates a broad class of functions, including additive and multiplicative Besov functions as special cases. These results advance the theoretical foundations of Bayesian neural networks and provide rigorous justification for their practical effectiveness in high-dimensional, structured estimation problems.
Authors: Benjamin Graham
Abstract: Image registration is used in many medical image analysis applications, such as tracking the motion of tissue in cardiac images, where cardiac kinematics can be an indicator of tissue health. Registration is a challenging problem for deep learning algorithms because ground truth transformations are not feasible to create, and because there are potentially multiple transformations that can produce images that appear correlated with the goal. Unsupervised methods have been proposed to learn to predict effective transformations, but these methods take significantly longer to predict than established baseline methods. For a deep learning method to see adoption in wider research and clinical settings, it should be designed to run in a reasonable time on common, mid-level hardware. Fast methods have been proposed for the task of image registration but often use patch-based methods which can affect registration accuracy for a highly dynamic organ such as the heart. In this thesis, a fast, volumetric registration model is proposed for the use of quantifying cardiac strain. The proposed Deep Learning Neural Network (DLNN) is designed to utilize an architecture that can compute convolutions incredibly efficiently, allowing the model to achieve registration fidelity similar to other state-of-the-art models while taking a fraction of the time to perform inference. The proposed fast and lightweight registration (FLIR) model is used to predict tissue motion which is then used to quantify the non-uniform strain experienced by the tissue. For acquisitions taken from the same patient at approximately the same time, it would be expected that strain values measured between the acquisitions would have very small differences. Using this metric, strain values computed using the FLIR method are shown to be very consistent.
Authors: Marc-Antoine Coulombe, Maxime Berger, Antoine Lesage-Landry
Abstract: The growing reliance on power electronics introduces new challenges requiring detailed time-domain analyses with fast and accurate circuit simulation tools. Currently, commercial time-domain simulation software are mainly relying on physics-based methods to simulate power electronics. Recent work showed that data-driven and physics-informed learning methods can increase simulation speed with limited compromise on accuracy, but many challenges remain before deployment in commercial tools can be possible. In this paper, we propose a physics-informed bidirectional long-short term memory neural network (BiLSTM-PINN) model to simulate the time-domain response of a closed-loop dc-dc boost converter for various operating points, parameters, and perturbations. A physics-informed fully-connected neural network (FCNN) and a BiLSTM are also trained to establish a comparison. The three methods are then compared using step-response tests to assess their performance and limitations in terms of accuracy. The results show that the BiLSTM-PINN and BiLSTM models outperform the FCNN model by more than 9 and 4.5 times, respectively, in terms of median RMSE. Their standard deviation values are more than 2.6 and 1.7 smaller than the FCNN's, making them also more consistent. Those results illustrate that the proposed BiLSTM-PINN is a potential alternative to other physics-based or data-driven methods for power electronics simulations.
Authors: Rasiq Hussain, Jerry Ma, Rithik Khandelwal, Joshua Oltmanns, Mehak Gupta
Abstract: Natural Language Processing (NLP) offers new avenues for personality assessment by leveraging rich, open-ended text, moving beyond traditional questionnaires. In this study, we address the challenge of modeling long narrative interview where each exceeds 2000 tokens so as to predict Five-Factor Model (FFM) personality traits. We propose a two-step approach: first, we extract contextual embeddings using sliding-window fine-tuning of pretrained language models; then, we apply Recurrent Neural Networks (RNNs) with attention mechanisms to integrate long-range dependencies and enhance interpretability. This hybrid method effectively bridges the strengths of pretrained transformers and sequence modeling to handle long-context data. Through ablation studies and comparisons with state-of-the-art long-context models such as LLaMA and Longformer, we demonstrate improvements in prediction accuracy, efficiency, and interpretability. Our results highlight the potential of combining language-based features with long-context modeling to advance personality assessment from life narratives.
Authors: Murtaza Rangwala, Richard O. Sinnott, Rajkumar Buyya
Abstract: Federated learning systems increasingly rely on diverse network topologies to address scalability and organizational constraints. While existing privacy research focuses on gradient-based attacks, the privacy implications of network topology knowledge remain critically understudied. We conduct the first comprehensive analysis of topology-based privacy leakage across realistic adversarial knowledge scenarios, demonstrating that adversaries with varying degrees of structural knowledge can infer sensitive data distribution patterns even under strong differential privacy guarantees. Through systematic evaluation of 4,720 attack instances, we analyze six distinct adversarial knowledge scenarios: complete topology knowledge and five partial knowledge configurations reflecting real-world deployment constraints. We propose three complementary attack vectors: communication pattern analysis, parameter magnitude profiling, and structural position correlation, achieving success rates of 84.1%, 65.0%, and 47.2% under complete knowledge conditions. Critically, we find that 80% of realistic partial knowledge scenarios maintain attack effectiveness above security thresholds, with certain partial knowledge configurations achieving performance superior to the baseline complete knowledge scenario. To address these vulnerabilities, we propose and empirically validate structural noise injection as a complementary defense mechanism across 808 configurations, demonstrating up to 51.4% additional attack reduction when properly layered with existing privacy techniques. These results establish that network topology represents a fundamental privacy vulnerability in federated learning systems while providing practical pathways for mitigation through topology-aware defense mechanisms.
Authors: Yuchang Zhu, Huazhen Zhong, Qunshu Lin, Haotong Wei, Xiaolong Sun, Zixuan Yu, Minghao Liu, Zibin Zheng, Liang Chen
Abstract: With the remarkable generative capabilities of large language models (LLMs), using LLM-generated data to train downstream models has emerged as a promising approach to mitigate data scarcity in specific domains and reduce time-consuming annotations. However, recent studies have highlighted a critical issue: iterative training on self-generated data results in model collapse, where model performance degrades over time. Despite extensive research on the implications of LLM-generated data, these works often neglect the importance of data diversity, a key factor in data quality. In this work, we aim to understand the implications of the diversity of LLM-generated data on downstream model performance. Specifically, we explore how varying levels of diversity in LLM-generated data affect downstream model performance. Additionally, we investigate the performance of models trained on data that mixes different proportions of LLM-generated data, which we refer to as synthetic data. Our experimental results show that, with minimal distribution shift, moderately diverse LLM-generated data can enhance model performance in scenarios with insufficient labeled data, whereas highly diverse generated data has a negative impact. We hope our empirical findings will offer valuable guidance for future studies on LLMs as data generators.
Authors: Timoteo Kelly, Abdulkadir Korkmaz, Samuel Mallet, Connor Souders, Sadra Aliakbarpour, Praveen Rao
Abstract: We present HARPT, a large-scale annotated corpus of mobile health app store reviews aimed at advancing research in user privacy and trust. The dataset comprises over 480,000 user reviews labeled into seven categories that capture critical aspects of trust in applications, trust in providers and privacy concerns. Creating HARPT required addressing multiple complexities, such as defining a nuanced label schema, isolating relevant content from large volumes of noisy data, and designing an annotation strategy that balanced scalability with accuracy. This strategy integrated rule-based filtering, iterative manual labeling with review, targeted data augmentation, and weak supervision using transformer-based classifiers to accelerate coverage. In parallel, a carefully curated subset of 7,000 reviews was manually annotated to support model development and evaluation. We benchmark a broad range of classification models, demonstrating that strong performance is achievable and providing a baseline for future research. HARPT is released as a public resource to support work in health informatics, cybersecurity, and natural language processing.
Authors: Haitao Huang, Chuangtao Chen, Qinglin Zhao
Abstract: The generation and preservation of complex quantum states against environmental noise are paramount challenges in advancing continuous-variable (CV) quantum information processing. This paper introduces a novel framework based on continuous-variable quantum diffusion principles, synergizing them with CV quantum neural networks (CVQNNs) to address these dual challenges. For the task of state generation, our Continuous-Variable Quantum Diffusion Generative model (CVQD-G) employs a physically driven forward diffusion process using a thermal loss channel, which is then inverted by a learnable, parameter-efficient backward denoising process based on a CVQNN with time-embedding. This framework's capability is further extended for state recovery by the Continuous-Variable Quantum Diffusion Restoration model (CVQD-R), a specialized variant designed to restore quantum states, particularly coherent states with unknown parameters, from thermal degradation. Extensive numerical simulations validate these dual capabilities, demonstrating the high-fidelity generation of diverse Gaussian (coherent, squeezed) and non-Gaussian (Fock, cat) states, typically with fidelities exceeding 99%, and confirming the model's ability to robustly restore corrupted states. Furthermore, a comprehensive complexity analysis reveals favorable training and inference costs, highlighting the framework's efficiency, scalability, and its potential as a robust tool for quantum state engineering and noise mitigation in realistic CV quantum systems.
Authors: Saad Qadeer, Panos Stinis, Hui Wan
Abstract: A long-standing obstacle in the use of machine-learnt surrogates with larger PDE systems is the onset of instabilities when solved numerically. Efforts towards ameliorating these have mostly concentrated on improving the accuracy of the surrogates or imbuing them with additional structure, and have garnered limited success. In this article, we study a prototype problem and draw insights that can help with more complex systems. In particular, we focus on a viscous Burgers'-ML system and, after identifying the cause of the instabilities, prescribe strategies to stabilize the coupled system. To improve the accuracy of the stabilized system, we next explore methods based on the Mori--Zwanzig formalism.
Authors: Hevish Cowlessur, Tansu Alpcan, Chandra Thapa, Seyit Camtepe, Neel Kanth Kundu
Abstract: Efficiently embedding high-dimensional datasets onto noisy and low-qubit quantum systems is a significant barrier to practical Quantum Machine Learning (QML). Approaches such as quantum autoencoders can be constrained by current hardware capabilities and may exhibit vulnerabilities to reconstruction attacks due to their invertibility. We propose Quantum Principal Geodesic Analysis (qPGA), a novel, non-invertible method for dimensionality reduction and qubit-efficient encoding. Executed classically, qPGA leverages Riemannian geometry to project data onto the unit Hilbert sphere, generating outputs inherently suitable for quantum amplitude encoding. This technique preserves the neighborhood structure of high-dimensional datasets within a compact latent space, significantly reducing qubit requirements for amplitude encoding. We derive theoretical bounds quantifying qubit requirements for effective encoding onto noisy systems. Empirical results on MNIST, Fashion-MNIST, and CIFAR-10 show that qPGA preserves local structure more effectively than both quantum and hybrid autoencoders. Additionally, we demonstrate that qPGA enhances resistance to reconstruction attacks due to its non-invertible nature. In downstream QML classification tasks, qPGA can achieve over 99% accuracy and F1-score on MNIST and Fashion-MNIST, outperforming quantum-dependent baselines. Initial tests on real hardware and noisy simulators confirm its potential for noise-resilient performance, offering a scalable solution for advancing QML applications.
Authors: Mihailo Stojnic
Abstract: We study classical asymmetric binary perceptron (ABP) and associated \emph{local entropy} (LE) as potential source of its algorithmic hardness. Isolation of \emph{typical} ABP solutions in SAT phase seemingly suggests a universal algorithmic hardness. Paradoxically, efficient algorithms do exist even for constraint densities $\alpha$ fairly close but at a finite distance (\emph{computational gap}) from the capacity. In recent years, existence of rare large dense clusters and magical ability of fast algorithms to find them have been posited as the conceptual resolution of this paradox. Monotonicity or breakdown of the LEs associated with such \emph{atypical} clusters are predicated to play a key role in their thinning-out or even complete defragmentation. Invention of fully lifted random duality theory (fl RDT) [90,93,94] allows studying random structures \emph{typical} features. A large deviation upgrade, sfl LD RDT [96,97], moves things further and enables \emph{atypical} features characterizations as well. Utilizing the machinery of [96,97] we here develop a generic framework to study LE as an ABP's atypical feature. Already on the second level of lifting we discover that the LE results are closely matching those obtained through replica methods. For classical zero threshold ABP, we obtain that LE breaks down for $\alpha$ in $(0.77,0.78)$ interval which basically matches $\alpha\sim 0.75-0.77$ range that currently best ABP solvers can handle and effectively indicates that LE's behavior might indeed be among key reflections of the ABP's computational gaps presumable existence.
Authors: Feiting Yang, Antoine Moevus, Steve L\'evesque
Abstract: Human-Computer Interaction (HCI) has evolved significantly to incorporate emotion recognition capabilities, creating unprecedented opportunities for adaptive and personalized user experiences. This paper explores the integration of emotion detection into calendar applications, enabling user interfaces to dynamically respond to users' emotional states and stress levels, thereby enhancing both productivity and engagement. We present and evaluate two complementary approaches to emotion detection: a biometric-based method utilizing heart rate (HR) data extracted from electrocardiogram (ECG) signals processed through Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) neural networks to predict the emotional dimensions of Valence, Arousal, and Dominance; and a behavioral method analyzing computer activity through multiple machine learning models to classify emotions based on fine-grained user interactions such as mouse movements, clicks, and keystroke patterns. Our comparative analysis, from real-world datasets, reveals that while both approaches demonstrate effectiveness, the computer activity-based method delivers superior consistency and accuracy, particularly for mouse-related interactions, which achieved approximately 90\% accuracy. Furthermore, GRU networks outperformed LSTM models in the biometric approach, with Valence prediction reaching 84.38\% accuracy.
Authors: Jiahui Hu, Wenjun Dong
Abstract: We present Compressible Atmospheric Model-Network (CAM-NET), an AI model designed to predict neutral atmospheric variables from the Earth's surface to the ionosphere with high accuracy and computational efficiency. Accurate modeling of the entire atmosphere is critical for understanding the upward propagation of gravity waves, which influence upper-atmospheric dynamics and coupling across atmospheric layers. CAM-NET leverages the Spherical Fourier Neural Operator (SFNO) to capture global-scale atmospheric dynamics while preserving the Earth's spherical structure. Trained on a decade of datasets from the Whole Atmosphere Community Climate Model with thermosphere and ionosphere eXtension (WACCM-X), CAM-NET demonstrates accuracy comparable to WACCM-X while achieving a speedup of over 1000x in inference time, can provide one year simulation within a few minutes once trained. The model effectively predicts key atmospheric parameters, including zonal and meridional winds, temperature, and time rate of pressure. Inspired by traditional modeling approaches that use external couplers to simulate tracer transport, CAM-NET introduces a modular architecture that explicitly separates tracer prediction from core dynamics. The core backbone of CAM-NET focuses on forecasting primary physical variables (e.g., temperature, wind velocity), while tracer variables are predicted through a lightweight, fine-tuned model. This design allows for efficient adaptation to specific tracer scenarios with minimal computational cost, avoiding the need to retrain the entire model. We have validated this approach on the $O^2$ tracer, demonstrating strong performance and generalization capabilities.
Authors: Ye Tian, Zhang Yumin, Yifan Jia, Jianguo Sun, Yanbin Wang
Abstract: URL+HTML feature fusion shows promise for robust malicious URL detection, since attacker artifacts persist in DOM structures. However, prior work suffers from four critical shortcomings: (1) incomplete URL modeling, failing to jointly capture lexical patterns and semantic context; (2) HTML graph sparsity, where threat-indicative nodes (e.g., obfuscated scripts) are isolated amid benign content, causing signal dilution during graph aggregation; (3) unidirectional analysis, ignoring URL-HTML feature bidirectional interaction; and (4) opaque decisions, lacking attribution to malicious DOM components. To address these challenges, we present WebGuard++, a detection framework with 4 novel components: 1) Cross-scale URL Encoder: Hierarchically learns local-to-global and coarse to fine URL features based on Transformer network with dynamic convolution. 2) Subgraph-aware HTML Encoder: Decomposes DOM graphs into interpretable substructures, amplifying sparse threat signals via Hierarchical feature fusion. 3) Bidirectional Coupling Module: Aligns URL and HTML embeddings through cross-modal contrastive learning, optimizing inter-modal consistency and intra-modal specificity. 4) Voting Module: Localizes malicious regions through consensus voting on malicious subgraph predictions. Experiments show WebGuard++ achieves significant improvements over state-of-the-art baselines, achieving 1.1x-7.9x higher TPR at fixed FPR of 0.001 and 0.0001 across both datasets.
Authors: Khuram Naveed, Bruna Neves de Freitas, Ruben Pauwels
Abstract: Convolutional denoising autoencoders (DAEs) are powerful tools for image restoration. However, they inherit a key limitation of convolutional neural networks (CNNs): they tend to recover low-frequency features, such as smooth regions, more effectively than high-frequency details. This leads to the loss of fine details, which is particularly problematic in dental radiographs where preserving subtle anatomical structures is crucial. While self-attention mechanisms can help mitigate this issue by emphasizing important features, conventional attention methods often prioritize features corresponding to cleaner regions and may overlook those obscured by noise. To address this limitation, we propose a noise-aware self-attention method, which allows the model to effectively focus on and recover key features even within noisy regions. Building on this approach, we introduce the noise-aware attention-enhanced denoising autoencoder (NAADA) network for enhancing noisy panoramic dental radiographs. Compared with the recent state of the art (and much heavier) methods like Uformer, MResDNN etc., our method improves the reconstruction of fine details, ensuring better image quality and diagnostic accuracy.
Authors: Seunghun Lee, Jihong Park, Jinho Choi, Hyuncheol Park
Abstract: Tokens are fundamental processing units of generative AI (GenAI) and large language models (LLMs), and token communication (TC) is essential for enabling remote AI-generate content (AIGC) and wireless LLM applications. Unlike traditional bits, each of which is independently treated, the semantics of each token depends on its surrounding context tokens. This inter-token dependency makes TC vulnerable to outage channels, where the loss of a single token can significantly distort the original message semantics. Motivated by this, this paper focuses on optimizing token packetization to maximize the average token similarity (ATS) between the original and received token messages under outage channels. Due to inter-token dependency, this token grouping problem is combinatorial, with complexity growing exponentially with message length. To address this, we propose a novel framework of semantic packet aggregation with lookahead search (SemPA-Look), built on two core ideas. First, it introduces the residual semantic score (RSS) as a token-level surrogate for the message-level ATS, allowing robust semantic preservation even when a certain token packet is lost. Second, instead of full search, SemPA-Look applies a lookahead search-inspired algorithm that samples intra-packet token candidates without replacement (fixed depth), conditioned on inter-packet token candidates sampled with replacement (fixed width), thereby achieving linear complexity. Experiments on a remote AIGC task with the MS-COCO dataset (text captioned images) demonstrate that SemPA-Look achieves high ATS and LPIPS scores comparable to exhaustive search, while reducing computational complexity by up to 40$\times$. Compared to other linear-complexity algorithms such as the genetic algorithm (GA), SemPA-Look achieves 10$\times$ lower complexity, demonstrating its practicality for remote AIGC and other TC applications.
Authors: Farnood Salehi, Vandit Sharma, Amirhossein Askari Farsangi, Tun\c{c} Ozan Ayd{\i}n
Abstract: Modern deep learning models in computer vision require large datasets of real images, which are difficult to curate and pose privacy and legal concerns, limiting their commercial use. Recent works suggest synthetic data as an alternative, yet models trained with it often underperform. This paper proposes a two-step approach to bridge this gap. First, we propose an improved neural fractal formulation through which we introduce a new class of synthetic data. Second, we propose reverse stylization, a technique that transfers visual features from a small, license-free set of real images onto synthetic datasets, enhancing their effectiveness. We analyze the domain gap between our synthetic datasets and real images using Kernel Inception Distance (KID) and show that our method achieves a significantly lower distributional gap compared to existing synthetic datasets. Furthermore, our experiments across different tasks demonstrate the practical impact of this reduced gap. We show that pretraining the EDM2 diffusion model on our synthetic dataset leads to an 11% reduction in FID during image generation, compared to models trained on existing synthetic datasets, and a 20% decrease in autoencoder reconstruction error, indicating improved performance in data representation. Furthermore, a ViT-S model trained for classification on this synthetic data achieves over a 10% improvement in ImageNet-100 accuracy. Our work opens up exciting possibilities for training practical models when sufficiently large real training sets are not available.
Authors: Yan Jiang, Hao Zhou, LiZhong GU, Ai Han, TianLong Li
Abstract: LLMs' reliance on static knowledge and fragile tool invocation severely hinders the orchestration of complex, heterogeneous toolchains, particularly at large scales. Existing methods typically use rigid single-path execution, resulting in poor error recovery and exponentially growing search spaces. We introduce NaviAgent, a graph-navigated bilevel planning architecture for robust function calling, comprising a Multi-Path Decider and Graph-Encoded Navigator. As an LLM-powered agent, the Multi-Path Decider defines a four-dimensional decision space and continuously perceives environmental states, dynamically selecting the optimal action to fully cover all tool invocation scenarios. The Graph-Encoded Navigator constructs a Tool Dependency Heterogeneous Graph (TDHG), where node embeddings explicitly fuse API schema structure with historical invocation behavior. It also integrates a novel heuristic search strategy that guides the Decider toward efficient and highly successful toolchains, even for unseen tool combinations. Experiments show that NaviAgent consistently achieves the highest task success rate (TSR) across all foundation models and task complexities, outperforming the average baselines (ReAct, ToolLLM, {\alpha}-UMI) by 13.5%, 16.4%, and 19.0% on Qwen2.5-14B, Qwen2.5-32B, and Deepseek-V3, respectively. Its execution steps are typically within one step of the most efficient baseline, ensuring a strong balance between quality and efficiency. Notably, a fine-tuned Qwen2.5-14B model achieves a TSR of 49.5%, surpassing the much larger 32B model (44.9%) under our architecture. Incorporating the Graph-Encoded Navigator further boosts TSR by an average of 2.4 points, with gains up over 9 points on complex tasks for larger models (Deepseek-V3 and GPT-4o), highlighting its essential role in toolchain orchestration.
Authors: Aleksandr Algazinov, Matt Laing, Paul Laban
Abstract: Accessibility remains a critical concern in today's society, as many technologies are not developed to support the full range of user needs. Existing multi-agent systems (MAS) often cannot provide comprehensive assistance for users in need due to the lack of customization stemming from closed-source designs. Consequently, individuals with disabilities frequently encounter significant barriers when attempting to interact with digital environments. We introduce MATE, a multimodal accessibility MAS, which performs the modality conversions based on the user's needs. The system is useful for assisting people with disabilities by ensuring that data will be converted to an understandable format. For instance, if the user cannot see well and receives an image, the system converts this image to its audio description. MATE can be applied to a wide range of domains, industries, and areas, such as healthcare, and can become a useful assistant for various groups of users. The system supports multiple types of models, ranging from LLM API calling to using custom machine learning (ML) classifiers. This flexibility ensures that the system can be adapted to various needs and is compatible with a wide variety of hardware. Since the system is expected to run locally, it ensures the privacy and security of sensitive information. In addition, the framework can be effectively integrated with institutional technologies (e.g., digital healthcare service) for real-time user assistance. Furthermore, we introduce ModCon-Task-Identifier, a model that is capable of extracting the precise modality conversion task from the user input. Numerous experiments show that ModCon-Task-Identifier consistently outperforms other LLMs and statistical models on our custom data. Our code and data are publicly available at https://github.com/AlgazinovAleksandr/Multi-Agent-MATE.
URLs: https://github.com/AlgazinovAleksandr/Multi-Agent-MATE.
Authors: Tao Huang, Zhekun Liu, Rui Wang, Yang Zhang, Liping Jing
Abstract: Despite the remarkable multimodal capabilities of Large Vision-Language Models (LVLMs), discrepancies often occur between visual inputs and textual outputs--a phenomenon we term visual hallucination. This critical reliability gap poses substantial risks in safety-critical Artificial Intelligence (AI) applications, necessitating a comprehensive evaluation benchmark and effective detection methods. Firstly, we observe that existing visual-centric hallucination benchmarks mainly assess LVLMs from a perception perspective, overlooking hallucinations arising from advanced reasoning capabilities. We develop the Perception-Reasoning Evaluation Hallucination (PRE-HAL) dataset, which enables the systematic evaluation of both perception and reasoning capabilities of LVLMs across multiple visual semantics, such as instances, scenes, and relations. Comprehensive evaluation with this new benchmark exposed more visual vulnerabilities, particularly in the more challenging task of relation reasoning. To address this issue, we propose, to the best of our knowledge, the first Dempster-Shafer theory (DST)-based visual hallucination detection method for LVLMs through uncertainty estimation. This method aims to efficiently capture the degree of conflict in high-level features at the model inference phase. Specifically, our approach employs simple mass functions to mitigate the computational complexity of evidence combination on power sets. We conduct an extensive evaluation of state-of-the-art LVLMs, LLaVA-v1.5, mPLUG-Owl2 and mPLUG-Owl3, with the new PRE-HAL benchmark. Experimental results indicate that our method outperforms five baseline uncertainty metrics, achieving average AUROC improvements of 4%, 10%, and 7% across three LVLMs. Our code is available at https://github.com/HT86159/Evidential-Conflict.
Authors: Ankita Raj, Ambar Pal, Chetan Arora
Abstract: Backdoor attacks embed a hidden functionality into deep neural networks, causing the network to display anomalous behavior when activated by a predetermined pattern in the input Trigger, while behaving well otherwise on public test data. Recent works have shown that backdoored face recognition (FR) systems can respond to natural-looking triggers like a particular pair of sunglasses. Such attacks pose a serious threat to the applicability of FR systems in high-security applications. We propose a novel technique to (1) detect whether an FR network is compromised with a natural, physically realizable trigger, and (2) identify such triggers given a compromised network. We demonstrate the effectiveness of our methods with a compromised FR network, where we are able to identify the trigger (e.g., green sunglasses or red hat) with a top-5 accuracy of 74%, whereas a naive brute force baseline achieves 56% accuracy.
Authors: Debabrata Mahapatra, Shubham Agarwal, Apoorv Saxena, Subrata Mitra
Abstract: Prior work on input-token importance in auto-regressive transformers has relied on Softmax-normalized attention weights, which obscure the richer structure of pre-Softmax query-key logits. We introduce RCStat, a statistical framework that harnesses raw attention logits via Relative Contextualization (RC), a random variable measuring contextual alignment between token segments, and derive an efficient upper bound for RC. We demonstrate two applications: (i) Key-Value compression, where RC-based thresholds drive adaptive key-value eviction for substantial cache reduction with minimal quality loss; and (ii) Attribution, where RC yields higher-fidelity token-, sentence-, and chunk-level explanations than post-Softmax methods. Across question answering, summarization, and attribution benchmarks, RCStat achieves significant empirical gains, delivering state-of-the-art compression and attribution performance without any model retraining.
Authors: Jakob Ambsdorf, Asbj{\o}rn Munk, Sebastian Llambias, Anders Nymark Christensen, Kamil Mikolaj, Randall Balestriero, Martin Tolsgaard, Aasa Feragen, Mads Nielsen
Abstract: With access to large-scale, unlabeled medical datasets, researchers are confronted with two questions: Should they attempt to pretrain a custom foundation model on this medical data, or use transfer-learning from an existing generalist model? And, if a custom model is pretrained, are novel methods required? In this paper we explore these questions by conducting a case-study, in which we train a foundation model on a large regional fetal ultrasound dataset of 2M images. By selecting the well-established DINOv2 method for pretraining, we achieve state-of-the-art results on three fetal ultrasound datasets, covering data from different countries, classification, segmentation, and few-shot tasks. We compare against a series of models pretrained on natural images, ultrasound images, and supervised baselines. Our results demonstrate two key insights: (i) Pretraining on custom data is worth it, even if smaller models are trained on less data, as scaling in natural image pretraining does not translate to ultrasound performance. (ii) Well-tuned methods from computer vision are making it feasible to train custom foundation models for a given medical domain, requiring no hyperparameter tuning and little methodological adaptation. Given these findings, we argue that a bias towards methodological innovation should be avoided when developing domain specific foundation models under common computational resource constraints.
Authors: Federico Tavella, Kathryn Mearns, Angelo Cangelosi
Abstract: Robotic scene understanding increasingly relies on vision-language models (VLMs) to generate natural language descriptions of the environment. In this work, we present a comparative study of captioning strategies for tabletop scenes captured by a robotic arm equipped with an RGB camera. The robot collects images of objects from multiple viewpoints, and we evaluate several models that generate scene descriptions. We compare the performance of various captioning models, like BLIP and VLMs. Our experiments examine the trade-offs between single-view and multi-view captioning, and difference between recognising real-world and 3D printed objects. We quantitatively evaluate object identification accuracy, completeness, and naturalness of the generated captions. Results show that VLMs can be used in robotic settings where common objects need to be recognised, but fail to generalise to novel representations. Our findings provide practical insights into deploying foundation models for embodied agents in real-world settings.
Authors: Lujun Li, Yiqun Wang, Radu State
Abstract: Cloud cover in multispectral imagery (MSI) poses significant challenges for early season crop mapping, as it leads to missing or corrupted spectral information. Synthetic aperture radar (SAR) data, which is not affected by cloud interference, offers a complementary solution, but lack sufficient spectral detail for precise crop mapping. To address this, we propose a novel framework, Time-series MSI Image Reconstruction using Vision Transformer (ViT), to reconstruct MSI data in cloud-covered regions by leveraging the temporal coherence of MSI and the complementary information from SAR from the attention mechanism. Comprehensive experiments, using rigorous reconstruction evaluation metrics, demonstrate that Time-series ViT framework significantly outperforms baselines that use non-time-series MSI and SAR or time-series MSI without SAR, effectively enhancing MSI image reconstruction in cloud-covered regions.
Authors: Zhiyuan Wang, Bokui Chen
Abstract: Continual learning (CL) empowers pre-trained vision-language models to adapt effectively to novel or previously underrepresented data distributions without comprehensive retraining, enhancing their adaptability and efficiency. While vision-language models like CLIP show great promise, they struggle to maintain performance across domains in incremental learning scenarios. Existing prompt learning methods face two main limitations: 1) they primarily focus on class-incremental learning scenarios, lacking specific strategies for multi-domain task incremental learning; 2) most current approaches employ single-modal prompts, neglecting the potential benefits of cross-modal information exchange. To address these challenges, we propose the \ChordPrompt framework, which facilitates a harmonious interplay between visual and textual prompts. \ChordPrompt introduces cross-modal prompts to leverage interactions between visual and textual information. Our approach also employs domain-adaptive text prompts to select appropriate prompts for continual adaptation across multiple domains. Comprehensive experiments on multi-domain incremental learning benchmarks demonstrate that \ChordPrompt outperforms state-of-the-art methods in zero-shot generalization and downstream task performance.
Authors: Leon Klein, Atharva Kelkar, Aleksander Durumeric, Yaoyi Chen, Frank No\'e
Abstract: Coarse-grained (CG) molecular dynamics simulations extend the length and time scale of atomistic simulations by replacing groups of correlated atoms with CG beads. Machine-learned coarse-graining (MLCG) has recently emerged as a promising approach to construct highly accurate force fields for CG molecular dynamics. However, the calibration of MLCG force fields typically hinges on force matching, which demands extensive reference atomistic trajectories with corresponding force labels. In practice, atomistic forces are often not recorded, making traditional force matching infeasible on pre-existing datasets. Recently, noise-based kernels have been introduced to adapt force matching to the low-data regime, including situations in which reference atomistic forces are not present. While this approach produces force fields which recapitulate slow collective motion, it introduces significant local distortions due to the corrupting effects of the noise-based kernel. In this work, we introduce more general kernels based on normalizing flows that substantially reduce these local distortions while preserving global conformational accuracy. We demonstrate our method on small proteins, showing that flow-based kernels can generate high-quality CG forces solely from configurational samples.
Authors: Letian Kang, Shixian Luo, Yiqiang Li, Xiaoyang Yu, Shenxuan Zhou, Yong Wu
Abstract: Vision-Language Models (VLMs) have demonstrated strong performance in video-language tasks, yet their application to long video understanding remains constrained by the quadratic complexity of standard attention mechanisms. In this paper, we propose \textbf{PEVLM}, a parallel encoding strategy specifically designed to improve the prefill efficiency of VLMs without requiring model finetuning. PEVLM partitions the input into block-wise segments with a shared sink, preserves full-attention positional embeddings, and aligns attention weights to mimic full-attention distributions. This design reduces attention computation from $O((T \times N)^2)$ to $O(T \times N)$ while maintaining high accuracy. Extensive experiments on the LongVideoBench benchmark show that PEVLM achieves up to 8.37\% accuracy improvement over existing inference-efficient methods and delivers up to 7.47x speedup in attention computation and 40\% reduction in end-to-end latency. Under strict latency constraints, PEVLM significantly outperforms baselines, raising accuracy from 23.26\% to 61.03\%. These results highlight PEVLM's effectiveness for low-latency, long-context video understanding, making it well-suited for real-world applications such as autonomous driving.
Authors: Maciej Besta, Shriram Chandran, Jakub Cudak, Patrick Iff, Marcin Copik, Robert Gerstenberger, Tomasz Szydlo, J\"urgen M\"uller, Torsten Hoefler
Abstract: Recent advances in graph databases (GDBs) have been driving interest in large-scale analytics, yet current systems fail to support higher-order (HO) interactions beyond first-order (one-hop) relations, which are crucial for tasks such as subgraph counting, polyadic modeling, and HO graph learning. We address this by introducing a new class of systems, higher-order graph databases (HO-GDBs) that use lifting and lowering paradigms to seamlessly extend traditional GDBs with HO. We provide a theoretical analysis of OLTP and OLAP queries, ensuring correctness, scalability, and ACID compliance. We implement a lightweight, modular, and parallelizable HO-GDB prototype that offers native support for hypergraphs, node-tuples, subgraphs, and other HO structures under a unified API. The prototype scales to large HO OLTP & OLAP workloads and shows how HO improves analytical tasks, for example enhancing accuracy of graph neural networks within a GDB by 44%. Our work ensures low latency and high query throughput, and generalizes both ACID-compliant and eventually consistent systems.
Authors: Xuesong Li, Dianye Huang, Yameng Zhang, Nassir Navab, Zhongliang Jiang
Abstract: Understanding medical ultrasound imaging remains a long-standing challenge due to significant visual variability caused by differences in imaging and acquisition parameters. Recent advancements in large language models (LLMs) have been used to automatically generate terminology-rich summaries orientated to clinicians with sufficient physiological knowledge. Nevertheless, the increasing demand for improved ultrasound interpretability and basic scanning guidance among non-expert users, e.g., in point-of-care settings, has not yet been explored. In this study, we first introduce the scene graph (SG) for ultrasound images to explain image content to ordinary and provide guidance for ultrasound scanning. The ultrasound SG is first computed using a transformer-based one-stage method, eliminating the need for explicit object detection. To generate a graspable image explanation for ordinary, the user query is then used to further refine the abstract SG representation through LLMs. Additionally, the predicted SG is explored for its potential in guiding ultrasound scanning toward missing anatomies within the current imaging view, assisting ordinary users in achieving more standardized and complete anatomical exploration. The effectiveness of this SG-based image explanation and scanning guidance has been validated on images from the left and right neck regions, including the carotid and thyroid, across five volunteers. The results demonstrate the potential of the method to maximally democratize ultrasound by enhancing its interpretability and usability for ordinaries.
Authors: Sjoerd Dirksen, Patrick Finke, Paul Geuchen, Dominik St\"oger, Felix Voigtlaender
Abstract: This paper studies the $\ell^p$-Lipschitz constants of ReLU neural networks $\Phi: \mathbb{R}^d \to \mathbb{R}$ with random parameters for $p \in [1,\infty]$. The distribution of the weights follows a variant of the He initialization and the biases are drawn from symmetric distributions. We derive high probability upper and lower bounds for wide networks that differ at most by a factor that is logarithmic in the network's width and linear in its depth. In the special case of shallow networks, we obtain matching bounds. Remarkably, the behavior of the $\ell^p$-Lipschitz constant varies significantly between the regimes $ p \in [1,2) $ and $ p \in [2,\infty] $. For $p \in [2,\infty]$, the $\ell^p$-Lipschitz constant behaves similarly to $\Vert g\Vert_{p'}$, where $g \in \mathbb{R}^d$ is a $d$-dimensional standard Gaussian vector and $1/p + 1/p' = 1$. In contrast, for $p \in [1,2)$, the $\ell^p$-Lipschitz constant aligns more closely to $\Vert g \Vert_{2}$.
Authors: Kristian Sotirov, Annie E. Paine, Savvas Varsamopoulos, Antonio A. Gentile, Osvaldo Simeone
Abstract: Offline model-based optimization (MBO) refers to the task of optimizing a black-box objective function using only a fixed set of prior input-output data, without any active experimentation. Recent work has introduced quantum extremal learning (QEL), which leverages the expressive power of variational quantum circuits to learn accurate surrogate functions by training on a few data points. However, as widely studied in the classical machine learning literature, predictive models may incorrectly extrapolate objective values in unexplored regions, leading to the selection of overly optimistic solutions. In this paper, we propose integrating QEL with conservative objective models (COM) - a regularization technique aimed at ensuring cautious predictions on out-of-distribution inputs. The resulting hybrid algorithm, COM-QEL, builds on the expressive power of quantum neural networks while safeguarding generalization via conservative modeling. Empirical results on benchmark optimization tasks demonstrate that COM-QEL reliably finds solutions with higher true objective values compared to the original QEL, validating its superiority for offline design problems.
Authors: Pola Bereta, Ioannis Diamantis
Abstract: Understanding temporal patterns in online search behavior is crucial for real-time marketing and trend forecasting. Google Trends offers a rich proxy for public interest, yet the high dimensionality and noise of its time-series data present challenges for effective clustering. This study evaluates three unsupervised clustering approaches, Symbolic Aggregate approXimation (SAX), enhanced SAX (eSAX), and Topological Data Analysis (TDA), applied to 20 Google Trends keywords representing major consumer categories. Our results show that while SAX and eSAX offer fast and interpretable clustering for stable time series, they struggle with volatility and complexity, often producing ambiguous ``catch-all'' clusters. TDA, by contrast, captures global structural features through persistent homology and achieves more balanced and meaningful groupings. We conclude with practical guidance for using symbolic and topological methods in consumer analytics and suggest that hybrid approaches combining both perspectives hold strong potential for future applications.
Authors: Yuqian Fu, Tinghong Chen, Jiajun Chai, Xihuai Wang, Songjun Tu, Guojun Yin, Wei Lin, Qichao Zhang, Yuanheng Zhu, Dongbin Zhao
Abstract: Large language models (LLMs) have achieved remarkable progress in reasoning tasks, yet the optimal integration of Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) remains a fundamental challenge. Through comprehensive analysis of token distributions, learning dynamics, and integration mechanisms from entropy-based perspectives, we reveal key differences between these paradigms: SFT induces coarse-grained global changes to LLM policy distributions, while RL performs fine-grained selective optimizations, with entropy serving as a critical indicator of training effectiveness. Building on these observations, we propose Supervised Reinforcement Fine-Tuning (SRFT), a single-stage method that unifies both fine-tuning paradigms through entropy-aware weighting mechanisms. Our approach simultaneously applies SFT and RL to directly optimize the LLM using demonstrations and self-exploration rollouts rather than through two-stage sequential methods. Extensive experiments show that SRFT achieves 59.1% average accuracy, outperforming zero-RL methods by 9.0% on five mathematical reasoning benchmarks and 10.9% on three out-of-distribution benchmarks.
Authors: Adrian Iulian Cristescu, Matteo Giordano
Abstract: Predicting the probability of default (PD) of prospective loans is a critical objective for financial institutions. In recent years, machine learning (ML) algorithms have achieved remarkable success across a wide variety of prediction tasks; yet, they remain relatively underutilised in credit risk analysis. This paper highlights the opportunities that ML algorithms offer to this field by comparing the performance of five predictive models-Random Forests, Decision Trees, XGBoost, Gradient Boosting and AdaBoost-to the predominantly used logistic regression, over a benchmark dataset from Scheule et al. (Credit Risk Analytics: The R Companion). Our findings underscore the strengths and weaknesses of each method, providing valuable insights into the most effective ML algorithms for PD prediction in the context of loan portfolios.
Authors: Yuqi Zhu, Yi Zhong, Jintian Zhang, Ziheng Zhang, Shuofei Qiao, Yujie Luo, Lun Du, Da Zheng, Huajun Chen, Ningyu Zhang
Abstract: Large Language Models (LLMs) hold promise in automating data analysis tasks, yet open-source models face significant limitations in these kinds of reasoning-intensive scenarios. In this work, we investigate strategies to enhance the data analysis capabilities of open-source LLMs. By curating a seed dataset of diverse, realistic scenarios, we evaluate models across three dimensions: data understanding, code generation, and strategic planning. Our analysis reveals three key findings: (1) Strategic planning quality serves as the primary determinant of model performance; (2) Interaction design and task complexity significantly influence reasoning capabilities; (3) Data quality demonstrates a greater impact than diversity in achieving optimal performance. We leverage these insights to develop a data synthesis methodology, demonstrating significant improvements in open-source LLMs' analytical reasoning capabilities.
Authors: Baochang Ren, Shuofei Qiao, Wenhao Yu, Huajun Chen, Ningyu Zhang
Abstract: Large Language Models (LLMs), particularly slow-thinking models, often exhibit severe hallucination, outputting incorrect content due to an inability to accurately recognize knowledge boundaries during reasoning. While Reinforcement Learning (RL) can enhance complex reasoning abilities, its outcome-oriented reward mechanism often lacks factual supervision over the thinking process, further exacerbating the hallucination problem. To address the high hallucination in slow-thinking models, we propose Knowledge-enhanced RL, KnowRL. KnowRL guides models to perform fact-based slow thinking by integrating a factuality reward, based on knowledge verification, into the RL training process, helping them recognize their knowledge boundaries. KnowRL guides models to perform fact-based slow thinking by integrating a factuality reward, based on knowledge verification, into the RL training process, helping them recognize their knowledge boundaries. This targeted factual input during RL training enables the model to learn and internalize fact-based reasoning strategies. By directly rewarding adherence to facts within the reasoning steps, KnowRL fosters a more reliable thinking process. Experimental results on three hallucination evaluation datasets and two reasoning evaluation datasets demonstrate that KnowRL effectively mitigates hallucinations in slow-thinking models while maintaining their original strong reasoning capabilities. Our code is available at https://github.com/zjunlp/KnowRL.
Authors: Felix Faltings, Hannes Stark, Regina Barzilay, Tommi Jaakkola
Abstract: We develop ProxelGen, a protein structure generative model that operates on 3D densities as opposed to the prevailing 3D point cloud representations. Representing proteins as voxelized densities, or proxels, enables new tasks and conditioning capabilities. We generate proteins encoded as proxels via a 3D CNN-based VAE in conjunction with a diffusion model operating on its latent space. Compared to state-of-the-art models, ProxelGen's samples achieve higher novelty, better FID scores, and the same level of designability as the training set. ProxelGen's advantages are demonstrated in a standard motif scaffolding benchmark, and we show how 3D density-based generation allows for more flexible shape conditioning.
Authors: Viatcheslav Gurev, Timothy Rumbell
Abstract: Sampling low-energy molecular conformations, spatial arrangements of atoms in a molecule, is a critical task for many different calculations performed in the drug discovery and optimization process. Numerous specialized equivariant networks have been designed to generate molecular conformations from 2D molecular graphs. Recently, non-equivariant transformer models have emerged as a viable alternative due to their capability to scale to improve generalization. However, the concern has been that non-equivariant models require a large model size to compensate the lack of equivariant bias. In this paper, we demonstrate that a well-chosen positional encoding effectively addresses these size limitations. A standard transformer model incorporating relative positional encoding for molecular graphs when scaled to 25 million parameters surpasses the current state-of-the-art non-equivariant base model with 64 million parameters on the GEOM-DRUGS benchmark. We implemented relative positional encoding as a negative attention bias that linearly increases with the shortest path distances between graph nodes at varying slopes for different attention heads, similar to ALiBi, a widely adopted relative positional encoding technique in the NLP domain. This architecture has the potential to serve as a foundation for a novel class of generative models for molecular conformations.
Authors: Saeed Mahloujifar, Chuan Guo, G. Edward Suh, Kamalika Chaudhuri
Abstract: Differential privacy (DP) has become the standard for private data analysis. Certain machine learning applications only require privacy protection for specific protected attributes. Using naive variants of differential privacy in such use cases can result in unnecessary degradation of utility. In this work, we refine the definition of DP to create a more general and flexible framework that we call feature differential privacy (FDP). Our definition is simulation-based and allows for both addition/removal and replacement variants of privacy, and can handle arbitrary and adaptive separation of protected and non-protected features. We prove the properties of FDP, such as adaptive composition, and demonstrate its implications for limiting attribute inference attacks. We also propose a modification of the standard DP-SGD algorithm that satisfies FDP while leveraging desirable properties such as amplification via sub-sampling. We apply our framework to various machine learning tasks and show that it can significantly improve the utility of DP-trained models when public features are available. For example, we train diffusion models on the AFHQ dataset of animal faces and observe a drastic improvement in FID compared to DP, from 286.7 to 101.9 at $\epsilon=8$, assuming that the blurred version of a training image is available as a public feature. Overall, our work provides a new approach to private data analysis that can help reduce the utility cost of DP while still providing strong privacy guarantees.
Authors: Susovan Pal, Praneeth Vepakomma
Abstract: The mean shift (MS) is a non-parametric, density-based, iterative algorithm that has prominent usage in clustering and image segmentation. A rigorous proof for its convergence in full generality remains unknown. Two significant steps in this direction were taken in the paper \cite{Gh1}, which proved that for \textit{sufficiently large bandwidth}, the MS algorithm with the Gaussian kernel always converges in any dimension, and also by the same author in \cite{Gh2}, proved that MS always converges in one dimension for kernels with differentiable, strictly decreasing, convex profiles. In the more recent paper \cite{YT}, they have proved the convergence in more generality,\textit{ without any restriction on the bandwidth}, with the assumption that the KDE $f$ has a continuous Lipschitz gradient on the closure of the convex hull of the trajectory of the iterated sequence of the mode estimate, and also satisfies the {\L}ojasiewicz property there. The main theoretical result of this paper is a generalization of those of \cite{Gh1}, where we show that (1) for\textit{ sufficiently large bandwidth} convergence is guaranteed in any dimension with \textit{any radially symmetric and strictly positive definite kernels}. The proof uses two alternate characterizations of radially symmetric positive definite smooth kernels by Schoenberg and Bernstein \cite{Fass}, and borrows some steps from the proofs in \cite{Gh1}. Although the authors acknowledge that the result in that paper is more restrictive than that of \cite{YT} due to the lower bandwidth limit, it uses a different set of assumptions than \cite{YT}, and the proof technique is different.
Authors: Vladislav Esaulov, M. Moein Esfahani
Abstract: We study NAFNet (Nonlinear Activation Free Network), a simple and efficient deep learning baseline for image restoration. By using CIFAR10 images corrupted with noise and blur, we conduct an ablation study of NAFNet's core components. Our baseline model implements SimpleGate activation, Simplified Channel Activation (SCA), and LayerNormalization. We compare this baseline to different variants that replace or remove components. Quantitative results (PSNR, SSIM) and examples illustrate how each modification affects restoration performance. Our findings support the NAFNet design: the SimpleGate and simplified attention mechanisms yield better results than conventional activations and attention, while LayerNorm proves to be important for stable training. We conclude with recommendations for model design, discuss potential improvements, and future work.
Authors: Xingyang Li, Muyang Li, Tianle Cai, Haocheng Xi, Shuo Yang, Yujun Lin, Lvmin Zhang, Songlin Yang, Jinbo Hu, Kelly Peng, Maneesh Agrawala, Ion Stoica, Kurt Keutzer, Song Han
Abstract: Recent advances in diffusion models have enabled high-quality video generation, but the additional temporal dimension significantly increases computational costs, making training and inference on long videos prohibitively expensive. In this paper, we identify a phenomenon we term Spatiotemporal Energy Decay in video diffusion models: post-softmax attention scores diminish as spatial and temporal distance between tokens increase, akin to the physical decay of signal or waves over space and time in nature. Motivated by this, we propose Radial Attention, a scalable sparse attention mechanism with $O(n \log n)$ complexity that translates energy decay into exponentially decaying compute density, which is significantly more efficient than standard $O(n^2)$ dense attention and more expressive than linear attention. Specifically, Radial Attention employs a simple, static attention mask where each token attends to spatially nearby tokens, with the attention window size shrinking with temporal distance. Moreover, it allows pre-trained video diffusion models to extend their generation length with efficient LoRA-based fine-tuning. Extensive experiments show that Radial Attention maintains video quality across Wan2.1-14B, HunyuanVideo, and Mochi 1, achieving up to a 1.9$\times$ speedup over the original dense attention. With minimal tuning, it enables video generation up to 4$\times$ longer while reducing training costs by up to 4.4$\times$ compared to direct fine-tuning and accelerating inference by up to 3.7$\times$ compared to dense attention inference.
Authors: Renato De Leone, Francesca Maggioni, Andrea Spinelli
Abstract: In this paper, we introduce novel Twin Parametric Margin Support Vector Machine (TPMSVM) models designed to address multiclass classification tasks under feature uncertainty. To handle data perturbations, we construct bounded-by-norm uncertainty set around each training observation and derive the robust counterparts of the deterministic models using robust optimization techniques. To capture complex data structure, we explore both linear and kernel-induced classifiers, providing computationally tractable reformulations of the resulting robust models. Additionally, we propose two alternatives for the final decision function, enhancing models' flexibility. Finally, we validate the effectiveness of the proposed robust multiclass TPMSVM methodology on real-world datasets, showing the good performance of the approach in the presence of uncertainty.
Authors: Lingkai Kong, Wenhao Mu, Jiaming Cui, Yuchen Zhuang, B. Aditya Prakash, Bo Dai, Chao Zhang
Abstract: Decision-focused learning (DFL), which differentiates through the KKT conditions, has recently emerged as a powerful approach for predict-then-optimize problems. However, under probabilistic settings, DFL faces three major bottlenecks: model mismatch error, sample average approximation error, and gradient approximation error. Model mismatch error stems from the misalignment between the model's parameterized predictive distribution and the true probability distribution. Sample average approximation error arises when using finite samples to approximate the expected optimization objective. Gradient approximation error occurs when the objectives are non-convex and KKT conditions cannot be directly applied. In this paper, we present DF2, the first distribution-free decision-focused learning method designed to mitigate these three bottlenecks. Rather than depending on a task-specific forecaster that requires precise model assumptions, our method directly learns the expected optimization function during training. To efficiently learn this function in a data-driven manner, we devise an attention-based model architecture inspired by the distribution-based parameterization of the expected objective. We evaluate DF2 on two synthetic problems and three real-world problems, demonstrating the effectiveness of DF2. Our code is available at: https://github.com/Lingkai-Kong/DF2.
Authors: Weiqiu You, Helen Qu, Marco Gatti, Bhuvnesh Jain, Eric Wong
Abstract: Self-attributing neural networks (SANNs) present a potential path towards interpretable models for high-dimensional problems, but often face significant trade-offs in performance. In this work, we formally prove a lower bound on errors of per-feature SANNs, whereas group-based SANNs can achieve zero error and thus high performance. Motivated by these insights, we propose Sum-of-Parts (SOP), a framework that transforms any differentiable model into a group-based SANN, where feature groups are learned end-to-end without group supervision. SOP achieves state-of-the-art performance for SANNs on vision and language tasks, and we validate that the groups are interpretable on a range of quantitative and semantic metrics. We further validate the utility of SOP explanations in model debugging and cosmological scientific discovery. Our code is available at https://github.com/BrachioLab/sop
Authors: Nicklas J\"averg{\aa}rd, Rainey Lyons, Adrian Muntean, Jonas Forsman
Abstract: We propose a method to generate statistically representative synthetic data from a given dataset. The main goal of our method is for the created data set to mimic the inter--feature correlations present in the original data, while also offering a tunable parameter to influence the privacy level. In particular, our method constructs a statistical map by using the empirical conditional distributions between the features of the original dataset. Part of the tunability is achieved by limiting the depths of conditional distributions that are being used. We describe in detail our algorithms used both in the construction of a statistical map and how to use this map to generate synthetic observations. This approach is tested in three different ways: with a hand calculated example; a manufactured dataset; and a real world energy-related dataset of consumption/production of households in Madeira Island. We evaluate the method by comparing the datasets using the Pearson correlation matrix with different levels of resolution and depths of correlation. These two considerations are being viewed as tunable parameters influencing the resulting datasets fidelity and privacy. The proposed methodology is general in the sense that it does not rely on the used test dataset. We expect it to be applicable in a much broader context than indicated here.
Authors: Mathieu Blondel, Vincent Roulet
Abstract: Artificial intelligence has recently experienced remarkable advances, fueled by large models, vast datasets, accelerated hardware, and, last but not least, the transformative power of differentiable programming. This new programming paradigm enables end-to-end differentiation of complex computer programs (including those with control flows and data structures), making gradient-based optimization of program parameters possible. As an emerging paradigm, differentiable programming builds upon several areas of computer science and applied mathematics, including automatic differentiation, graphical models, optimization and statistics. This book presents a comprehensive review of the fundamental concepts useful for differentiable programming. We adopt two main perspectives, that of optimization and that of probability, with clear analogies between the two. Differentiable programming is not merely the differentiation of programs, but also the thoughtful design of programs intended for differentiation. By making programs differentiable, we inherently introduce probability distributions over their execution, providing a means to quantify the uncertainty associated with program outputs.
Authors: Daniel Potts, Laura Weidensager
Abstract: We propose two algorithms for boosting random Fourier feature models for approximating high-dimensional functions. These methods utilize the classical and generalized analysis of variance (ANOVA) decomposition to learn low-order functions, where there are few interactions between the variables. Our algorithms are able to find an index set of important input variables and variable interactions reliably. Furthermore, we generalize already existing random Fourier feature models to an ANOVA setting, where terms of different order can be used. Our algorithms have the advantage of interpretability, meaning that the influence of every input variable is known in the learned model, even for dependent input variables. We give theoretical as well as numerical results that our algorithms perform well for sensitivity analysis. The ANOVA-boosting step reduces the approximation error of existing methods significantly.
Authors: Chao Chen, Chenghua Guo, Rui Xu, Xiangwen Liao, Xi Zhang, Sihong Xie, Hui Xiong, Philip Yu
Abstract: Graphical models have demonstrated their exceptional capabilities across numerous applications, such as social networks, citation networks, and online recommendation systems. However, their performance, confidence, and trustworthiness are often limited by the inherent randomness in data and the challenges of accurately modeling real-world complexities. There has been increased interest in developing uncertainty quantification (UQ) techniques tailored to graphical models. In this survey, we comprehensively examine existing works on UQ for graphical models, focusing on key aspects such as the sources, representation, handling, and evaluation of uncertainty. This survey distinguishes itself from most existing UQ surveys by specifically concentrating on UQ in graphical models, including probabilistic graphical models (PGMs) and graph neural networks (GNNs). After reviewing sources of uncertainty, we organize the work using two high-level dimensions: uncertainty representation and uncertainty handling. By offering a comprehensive overview of the current landscape, including both established methodologies and emerging trends, we aim to bridge gaps in understanding key challenges and opportunities in UQ for graphical models, hoping to inspire researchers working on graphical models or uncertainty quantification to make further advancements at the cross of the two fields.
Authors: Sonny Achten, Zander Op de Beeck, Francesco Tonin, Volkan Cevher, Johan A. K. Suykens
Abstract: Clustering nodes in heterophilous graphs is challenging as traditional methods assume that effective clustering is characterized by high intra-cluster and low inter-cluster connectivity. To address this, we introduce HeNCler-a novel approach for Heterophilous Node Clustering. HeNCler learns a similarity graph by optimizing a clustering-specific objective based on weighted kernel singular value decomposition. Our approach enables spectral clustering on an asymmetric similarity graph, providing flexibility for both directed and undirected graphs. By solving the primal problem directly, our method overcomes the computational difficulties of traditional adjacency partitioning-based approaches. Experimental results show that HeNCler significantly improves node clustering performance in heterophilous graph settings, highlighting the advantage of its asymmetric graph-learning framework.
Authors: Zahraa Al Sahili, Ioannis Patras, Matthew Purver
Abstract: Multimodal machine learning (MML) is rapidly reshaping the way mental-health disorders are detected, characterized, and longitudinally monitored. Whereas early studies relied on isolated data streams -- such as speech, text, or wearable signals -- recent research has converged on architectures that integrate heterogeneous modalities to capture the rich, complex signatures of psychiatric conditions. This survey provides the first comprehensive, clinically grounded synthesis of MML for mental health. We (i) catalog 26 public datasets spanning audio, visual, physiological signals, and text modalities; (ii) systematically compare transformer, graph, and hybrid-based fusion strategies across 28 models, highlighting trends in representation learning and cross-modal alignment. Beyond summarizing current capabilities, we interrogate open challenges: data governance and privacy, demographic and intersectional fairness, evaluation explainability, and the complexity of mental health disorders in multimodal settings. By bridging methodological innovation with psychiatric utility, this survey aims to orient both ML researchers and mental-health practitioners toward the next generation of trustworthy, multimodal decision-support systems.
Authors: Peter A. Wijeratne, Daniel C. Alexander
Abstract: Disease progression models infer group-level temporal trajectories of change in patients' features as a chronic degenerative condition plays out. They provide unique insight into disease biology and staging systems with individual-level clinical utility. Discrete models consider disease progression as a latent permutation of events, where each event corresponds to a feature becoming measurably abnormal. However, permutation inference using traditional maximum likelihood approaches becomes prohibitive due to combinatoric explosion, severely limiting model dimensionality and utility. Here we leverage ideas from optimal transport to model disease progression as a latent permutation matrix of events belonging to the Birkhoff polytope, facilitating fast inference via optimisation of the variational lower bound. This enables a factor of 1000 times faster inference than the current state of the art and, correspondingly, supports models with several orders of magnitude more features than the current state of the art can consider. Experiments demonstrate the increase in speed, accuracy and robustness to noise in simulation. Further experiments with real-world imaging data from two separate datasets, one from Alzheimer's disease patients, the other age-related macular degeneration, showcase, for the first time, pixel-level disease progression events in the brain and eye, respectively. Our method is low compute, interpretable and applicable to any progressive condition and data modality, giving it broad potential clinical utility.
Authors: Rui Sun, Zhipeng Wang, Hengrui Zhang, Ming Jiang, Yizhe Wen, Jiahao Sun, Xinyu Qu, Kezhi Li
Abstract: One of the biggest challenges of building artificial intelligence (AI) model in the healthcare area is the data sharing. Since healthcare data is private, sensitive, and heterogeneous, collecting sufficient data for modelling is exhausting, costly, and sometimes impossible. In this paper, we propose a framework for global healthcare modelling using datasets from multi-continents (Europe, North America, and Asia) without sharing the local datasets, and choose glucose management as a study model to verify its effectiveness. Technically, blockchain-enabled federated learning is implemented with adaptation to meet the privacy and safety requirements of healthcare data, meanwhile, it rewards honest participation and penalizes malicious activities using its on-chain incentive mechanism. Experimental results show that the proposed framework is effective, efficient, and privacy-preserving. Its prediction accuracy consistently outperforms models trained on limited personal data and achieves comparable or even slightly better results than centralized training in certain scenarios, all while preserving data privacy. This work paves the way for international collaborations on healthcare projects, where additional data is crucial for reducing bias and providing benefits to humanity.
Authors: Jesse He, Helen Jenne, Herman Chau, Davis Brown, Mark Raugas, Sara Billey, Henry Kvinge
Abstract: Machine learning is becoming an increasingly valuable tool in mathematics, enabling one to identify subtle patterns across collections of examples so vast that they would be impossible for a single researcher to feasibly review and analyze. In this work, we use graph neural networks to investigate \emph{quiver mutation} -- an operation that transforms one quiver (or directed multigraph) into another -- which is central to the theory of cluster algebras with deep connections to geometry, topology, and physics. In the study of cluster algebras, the question of \emph{mutation equivalence} is of fundamental concern: given two quivers, can one efficiently determine if one quiver can be transformed into the other through a sequence of mutations? In this paper, we use graph neural networks and AI explainability techniques to independently discover mutation equivalence criteria for quivers of type $\tilde{D}$. Along the way, we also show that even without explicit training to do so, our model captures structure within its hidden representation that allows us to reconstruct known criteria from type $D$, adding to the growing evidence that modern machine learning models are capable of learning abstract and parsimonious rules from mathematical data.
Authors: Gaurav Menghani, Ravi Kumar, Sanjiv Kumar
Abstract: One of the core pillars of efficient deep learning methods is architectural improvements such as the residual/skip connection, which has led to significantly better model convergence and quality. Since then the residual connection has become ubiquitous in not just convolutional neural networks but also transformer-based architectures, the backbone of LLMs. In this paper we introduce Learned Augmented Residual Layer (LAuReL) -- a novel generalization of the canonical residual connection -- with the goal to be an in-situ replacement of the latter while outperforming on both model quality and footprint metrics. Our experiments show that using LAuReL can help boost performance for both vision and language models. For example, on the ResNet-50, ImageNet 1K task, it achieves 60% of the gains from adding an extra layer, while only adding 0.003% more parameters, and matches it while adding 2.6 times fewer parameters. Similarly, when pre-training 1B and 4B parameter LLMs, LAuReL improves performance on a variety of challenging downstream evaluation tasks by 2.54% to 20.05%, while adding only 0.012% and 0.1% additional parameters, respectively.
Authors: Andrii Skliar, Ties van Rozendaal, Romain Lepert, Todor Boinovski, Mart van Baalen, Markus Nagel, Paul Whatmough, Babak Ehteshami Bejnordi
Abstract: Mixture of Experts (MoE) LLMs have recently gained attention for their ability to enhance performance by selectively engaging specialized subnetworks or "experts" for each input. However, deploying MoEs on memory-constrained devices remains challenging, particularly when generating tokens sequentially with a batch size of one, as opposed to typical high-throughput settings involving long sequences or large batches. In this work, we optimize MoE on memory-constrained devices where only a subset of expert weights fit in DRAM. We introduce a novel cache-aware routing strategy that leverages expert reuse during token generation to improve cache locality. We evaluate our approach on language modeling, MMLU, and GSM8K benchmarks and present on-device results demonstrating 2$\times$ speedups on mobile devices, offering a flexible, training-free solution to extend MoE's applicability across real-world applications.
Authors: Margaux Tornqvist, Jean-Daniel Zucker, Tristan Fauvel, Nicolas Lambert, Mathilde Berthelot, Antoine Movschin
Abstract: Access to large-scale high-quality healthcare databases is key to accelerate medical research and make insightful discoveries about diseases. However, access to such data is often limited by patient privacy concerns, data sharing restrictions and high costs. To overcome these limitations, synthetic patient data has emerged as an alternative. However, synthetic data generation (SDG) methods typically rely on machine learning (ML) models trained on original data, leading back to the data scarcity problem. We propose an approach to generate synthetic tabular patient data that does not require access to the original data, but only a description of the desired database. We leverage prior medical knowledge and in-context learning capabilities of large language models (LLMs) to generate realistic patient data, even in a low-resource setting. We quantitatively evaluate our approach against state-of-the-art SDG models, using fidelity, privacy, and utility metrics. Our results show that while LLMs may not match the performance of state-of-the-art models trained on the original data, they effectively generate realistic patient data with well-preserved clinical correlations. An ablation study highlights key elements of our prompt contributing to high-quality synthetic patient data generation. This approach, which is easy to use and does not require original data or advanced ML skills, is particularly valuable for quickly generating custom-designed patient data, supporting project implementation and providing educational resources.
Authors: Yeonhong Park, Jake Hyun, Hojoon Kim, Jae W. Lee
Abstract: Quantization of Large Language Models (LLMs) has recently gained popularity, particularly for on-device settings with limited hardware resources. While efficient, quantization inevitably degrades model quality, especially in aggressive low-bit settings such as 3-bit and 4-bit precision. In this paper, we propose DecDEC, an inference scheme that improves the quality of low-bit LLMs while preserving the key benefits of quantization: GPU memory savings and latency reduction. DecDEC stores the residual matrix -- the difference between full-precision and quantized weights -- in CPU, and dynamically fetches the residuals for only a small portion of the weights. This portion corresponds to the salient channels, marked by activation outliers, with the fetched residuals helping to correct quantization errors in these channels. Salient channels are identified dynamically at each decoding step by analyzing the input activations -- this enables adaptation to the dynamic nature of activation distribution, thus maximizing the effectiveness of error compensation. We demonstrate the effectiveness of DecDEC by augmenting state-of-the-art quantization methods. For example, DecDEC reduces the perplexity of a 3-bit Llama-3-8B-Instruct model from 10.15 to 9.12 -- outperforming its 3.5-bit counterpart -- while adding less than 0.0003\% to GPU memory usage and incurring only a 1.7\% inference slowdown on NVIDIA RTX 4050 Mobile.
Authors: Shadab Ahamed, Niloufar Zakariaei, Eldad Haber, Moshe Eliasof
Abstract: Training convolutional neural networks (CNNs) on high-resolution images is often bottlenecked by the cost of evaluating gradients of the loss on the finest spatial mesh. To address this, we propose Multiscale Gradient Estimation (MGE), a Multilevel Monte Carlo-inspired estimator that expresses the expected gradient on the finest mesh as a telescopic sum of gradients computed on progressively coarser meshes. By assigning larger batches to the cheaper coarse levels, MGE achieves the same variance as single-scale stochastic gradient estimation while reducing the number of fine mesh convolutions by a factor of 4 with each downsampling. We further embed MGE within a Full-Multiscale training algorithm that solves the learning problem on coarse meshes first and "hot-starts" the next finer level, cutting the required fine mesh iterations by an additional order of magnitude. Extensive experiments on image denoising, deblurring, inpainting and super-resolution tasks using UNet, ResNet and ESPCN backbones confirm the practical benefits: Full-Multiscale reduces the computation costs by 4-16$\times$ with no significant loss in performance. Together, MGE and Full-Multiscale offer a principled, architecture-agnostic route to accelerate CNN training on high-resolution data without sacrificing accuracy, and they can be combined with other variance-reduction or learning-rate schedules to further enhance scalability.
Authors: Hongyi Liu, Rajarshi Saha, Zhen Jia, Youngsuk Park, Jiaji Huang, Shoham Sabach, Yu-Xiang Wang, George Karypis
Abstract: Large Language Models (LLMs) have demonstrated exceptional performance in natural language processing tasks, yet their massive size makes serving them inefficient and costly. Semi-structured pruning has emerged as an effective method for model acceleration, but existing approaches are suboptimal because they focus on local, layer-wise optimizations using heuristic rules, failing to leverage global feedback. We present ProxSparse, a learning-based framework for mask selection enabled by regularized optimization. ProxSparse transforms the rigid, non-differentiable mask selection process into a smoother optimization procedure, allowing gradual mask exploration with flexibility. ProxSparse does not involve additional weight updates once the mask is determined. Our extensive evaluations on 7 widely used models show that ProxSparse consistently outperforms previously proposed semi-structured mask selection methods with significant improvement, demonstrating the effectiveness of our learned approach towards semi-structured pruning.
Authors: Sagi Meir, Tommer D. Keidar, Shlomi Reuveni, Barak Hirshberg
Abstract: Machine learning models have become indispensable tools in applications across the physical sciences. Their training is often time-consuming, vastly exceeding the inference timescales. Several protocols have been developed to perturb the learning process and improve the training, such as shrink and perturb, warm restarts, and stochastic resetting. For classifiers, these perturbations have been shown to result in enhanced speedups or improved generalization. However, the design of such perturbations is usually done ad hoc by intuition and trial and error. To rationally optimize training protocols, we frame them as first-passage processes and consider their response to perturbations. We show that if the unperturbed learning process reaches a quasi-steady state, the response at a single perturbation frequency can predict the behavior at a wide range of frequencies. We employ this approach to a CIFAR-10 classifier using the ResNet-18 model and identify a useful perturbation and frequency among several possibilities. We demonstrate the transferability of the approach to other datasets, architectures, optimizers and even tasks (regression instead of classification). Our work allows optimization of perturbations for improving the training of machine learning models using a first-passage approach.
Authors: Ayush K. Varshney, Vicen\c{c} Torra
Abstract: Machine Unlearning allows participants to remove their data from a trained machine learning model in order to preserve their privacy, and security. However, the machine unlearning literature for generative models is rather limited. The literature for image-to-image generative model (I2I model) considers minimizing the distance between Gaussian noise and the output of I2I model for forget samples as machine unlearning. However, we argue that the machine learning model performs fairly well on unseen data i.e., a retrained model will be able to catch generic patterns in the data and hence will not generate an output which is equivalent to Gaussian noise. In this paper, we consider that the model after unlearning should treat forget samples as out-of-distribution (OOD) data, i.e., the unlearned model should no longer recognize or encode the specific patterns found in the forget samples. To achieve this, we propose a framework which decouples the model parameters with gradient ascent, ensuring that forget samples are OOD for unlearned model with theoretical guarantee. We also provide $(\epsilon, \delta)$-unlearning guarantee for model updates with gradient ascent. The unlearned model is further fine-tuned on the remaining samples to maintain its performance. We also propose an attack model to ensure that the unlearned model has effectively removed the influence of forget samples. Extensive empirical evaluation on two large-scale datasets, ImageNet-1K and Places365 highlights the superiority of our approach. To show comparable performance with retrained model, we also show the comparison of a simple AutoEncoder on various baselines on CIFAR-10 dataset.
Authors: Riccardo Zamboni, Mirco Mutti, Marcello Restelli
Abstract: In reinforcement learning, we typically refer to unsupervised pre-training when we aim to pre-train a policy without a priori access to the task specification, i.e. rewards, to be later employed for efficient learning of downstream tasks. In single-agent settings, the problem has been extensively studied and mostly understood. A popular approach, called task-agnostic exploration, casts the unsupervised objective as maximizing the entropy of the state distribution induced by the agent's policy, from which principles and methods follow. In contrast, little is known about it in multi-agent settings, which are ubiquitous in the real world. What are the pros and cons of alternative problem formulations in this setting? How hard is the problem in theory, how can we solve it in practice? In this paper, we address these questions by first characterizing those alternative formulations and highlighting how the problem, even when tractable in theory, is non-trivial in practice. Then, we present a scalable, decentralized, trust-region policy search algorithm to address the problem in practical settings. Finally, we provide numerical validations to both corroborate the theoretical findings and pave the way for unsupervised multi-agent reinforcement learning via task-agnostic exploration in challenging domains, showing that optimizing for a specific objective, namely mixture entropy, provides an excellent trade-off between tractability and performances.
Authors: Alireza Kazemipour, Simone Parisi, Matthew E. Taylor, Michael Bowling
Abstract: A tenet of reinforcement learning is that the agent always observes rewards. However, this is not true in many realistic settings, e.g., a human observer may not always be available to provide rewards, sensors may be limited or malfunctioning, or rewards may be inaccessible during deployment. Monitored Markov decision processes (Mon-MDPs) have recently been proposed to model such settings. However, existing Mon-MDP algorithms have several limitations: they do not fully exploit the problem structure, cannot leverage a known monitor, lack worst-case guarantees for 'unsolvable' Mon-MDPs without specific initialization, and offer only asymptotic convergence proofs. This paper makes three contributions. First, we introduce a model-based algorithm for Mon-MDPs that addresses these shortcomings. The algorithm employs two instances of model-based interval estimation: one to ensure that observable rewards are reliably captured, and another to learn the minimax-optimal policy. Second, we empirically demonstrate the advantages. We show faster convergence than prior algorithms in over four dozen benchmarks, and even more dramatic improvement when the monitoring process is known. Third, we present the first finite-sample bound on performance. We show convergence to a minimax-optimal policy even when some rewards are never observable.
Authors: Dahun Shin, Dongyeop Lee, Jinseok Chung, Namhoon Lee
Abstract: Approximate second-order optimization methods often exhibit poorer generalization compared to first-order approaches. In this work, we look into this issue through the lens of the loss landscape and find that existing second-order methods tend to converge to sharper minima compared to SGD. In response, we propose Sassha, a novel second-order method designed to enhance generalization by explicitly reducing sharpness of the solution, while stabilizing the computation of approximate Hessians along the optimization trajectory. In fact, this sharpness minimization scheme is crafted also to accommodate lazy Hessian updates, so as to secure efficiency besides flatness. To validate its effectiveness, we conduct a wide range of standard deep learning experiments where Sassha demonstrates its outstanding generalization performance that is comparable to, and mostly better than, other methods. We provide a comprehensive set of analyses including convergence, robustness, stability, efficiency, and cost.
Authors: Shijun Zhang, Hongkai Zhao, Yimin Zhong, Haomin Zhou
Abstract: The architecture of a neural network and the selection of its activation function are both fundamental to its performance. Equally vital is ensuring these two elements are well-matched, as their alignment is key to achieving effective representation and learning. In this paper, we introduce the Fourier Multi-Component and Multi-Layer Neural Network (FMMNN), a novel model that creates a strong synergy between them. We demonstrate that FMMNNs are highly effective and flexible in modeling high-frequency components. Our theoretical results demonstrate that FMMNNs have exponential expressive power for function approximation. We also analyze the optimization landscape of FMMNNs and find it to be much more favorable than that of standard fully connected neural networks, especially when dealing with high-frequency features. In addition, we propose a scaled random initialization method for the first layer's weights in FMMNNs, which significantly speeds up training and enhances overall performance. Extensive numerical experiments support our theoretical insights, showing that FMMNNs consistently outperform traditional approaches in accuracy and efficiency across various tasks.
Authors: Ben Keslaki
Abstract: Stochastic Gradient Descent (SGD) and its variants, such as ADAM, are foundational to deep learning optimization, adjusting model parameters through fixed or adaptive learning rates based on loss function gradients. However, these methods often struggle to balance adaptability and efficiency in high-dimensional, non-convex settings. This paper introduces AYLA, a novel optimization framework that enhances training dynamics via loss function transformation. AYLA applies a tunable power-law transformation to the loss, preserving critical points while scaling loss values to amplify gradient sensitivity and accelerate convergence. Additionally, we propose an effective learning rate that dynamically adapts to the transformed loss, further improving optimization efficiency. Empirical evaluations on minimizing a synthetic non-convex polynomial, solving a non-convex curve-fitting task, and performing digit classification (MNIST) and image recognition (CIFAR-100) demonstrate that AYLA consistently outperforms SGD and ADAM in both convergence speed and training stability. By reshaping the loss landscape, AYLA provides a model-agnostic enhancement to existing optimization methods, offering a promising advancement in deep neural network training.
Authors: Senwei Liang, Chunmei Wang, Xingjian Xu
Abstract: Modeling stochastic differential equations (SDEs) is crucial for understanding complex dynamical systems in various scientific fields. Recent methods often employ neural network-based models, which typically represent SDEs through a combination of deterministic and stochastic terms. However, these models usually lack interpretability and have difficulty generalizing beyond their training domain. This paper introduces the Finite Expression Method (FEX), a symbolic learning approach designed to derive interpretable mathematical representations of the deterministic component of SDEs. For the stochastic component, we integrate FEX with advanced generative modeling techniques to provide a comprehensive representation of SDEs. The numerical experiments on linear, nonlinear, and multidimensional SDEs demonstrate that FEX generalizes well beyond the training domain and delivers more accurate long-term predictions compared to neural network-based methods. The symbolic expressions identified by FEX not only improve prediction accuracy but also offer valuable scientific insights into the underlying dynamics of the systems, paving the way for new scientific discoveries.
Authors: Jonathan Adam Rico, Nagarajan Raghavan, Senthilnath Jayavelu
Abstract: Fault diagnosis prevents train disruptions by ensuring the stability and reliability of their transmission systems. Data-driven fault diagnosis models have several advantages over traditional methods in terms of dealing with non-linearity, adaptability, scalability, and automation. However, existing data-driven models are trained on separate transmission components and only consider single faults due to the limitations of existing datasets. These models will perform worse in scenarios where components operate with each other at the same time, affecting each component's vibration signals. To address some of these challenges, we propose a frequency domain representation and a 1-dimensional convolutional neural network for compound fault diagnosis and applied it on the PHM Beijing 2024 dataset, which includes 21 sensor channels, 17 single faults, and 42 compound faults from 4 interacting components, that is, motor, gearbox, left axle box, and right axle box. Our proposed model achieved 97.67% and 93.93% accuracies on the test set with 17 single faults and on the test set with 42 compound faults, respectively.
Authors: Muhammad Haseeb Aslam, Clara Martinez, Marco Pedersoli, Alessandro Koerich, Ali Etemad, Eric Granger
Abstract: Advances in self-distillation have shown that when knowledge is distilled from a teacher to a student using the same deep learning (DL) architecture, the student performance can surpass the teacher particularly when the network is overparameterized and the teacher is trained with early stopping. Alternatively, ensemble learning also improves performance, although training, storing, and deploying multiple models becomes impractical as the number of models grows. Even distilling an ensemble to a single student model or weight averaging methods first requires training of multiple teacher models and does not fully leverage the inherent stochasticity for generating and distilling diversity in DL models. These constraints are particularly prohibitive in resource-constrained or latency-sensitive applications such as wearable devices. This paper proposes to train only one model and generate multiple diverse teacher representations using distillation-time dropout. However, generating these representations stochastically leads to noisy representations that are misaligned with the learned task. To overcome this problem, a novel stochastic self-distillation (SSD) training strategy is introduced for filtering and weighting teacher representation to distill from task-relevant representations only, using student-guided knowledge distillation (SGKD). The student representation at each distillation step is used as authority to guide the distillation process. Experimental results on real-world affective computing, wearable/biosignal datasets from the UCR Archive, the HAR dataset, and image classification datasets show that the proposed SSD method can outperform state-of-the-art methods without increasing the model size at both training and testing time, and incurs negligible computational complexity compared to state-of-the-art ensemble learning and weight averaging methods.
Authors: Muhammad Khalifa, Rishabh Agarwal, Lajanugen Logeswaran, Jaekyeom Kim, Hao Peng, Moontae Lee, Honglak Lee, Lu Wang
Abstract: Step-by-step verifiers -- also known as process reward models (PRMs) -- are a key ingredient for test-time scaling. PRMs require step-level supervision, making them expensive to train. This work aims to build data-efficient PRMs as verbalized step-wise reward models that verify every step in the solution by generating a verification chain-of-thought (CoT). We propose ThinkPRM, a long CoT verifier fine-tuned on orders of magnitude fewer process labels than those required by discriminative PRMs. Our approach capitalizes on the inherent reasoning abilities of long CoT models, and outperforms LLM-as-a-Judge and discriminative verifiers -- using only 1% of the process labels in PRM800K -- across several challenging benchmarks. Specifically, ThinkPRM beats the baselines on ProcessBench, MATH-500, and AIME '24 under best-of-N selection and reward-guided search. In an out-of-domain evaluation on a subset of GPQA-Diamond and LiveCodeBench, our PRM surpasses discriminative verifiers trained on the full PRM800K by 8% and 4.5%, respectively. Lastly, under the same token budget, ThinkPRM scales up verification compute more effectively compared to LLM-as-a-Judge, outperforming it by 7.2% on a subset of ProcessBench. Our work highlights the value of generative, long CoT PRMs that can scale test-time compute for verification while requiring minimal supervision for training. Our code, data, and models will be released at https://github.com/mukhal/thinkprm.
Authors: Thomas Flinkow, Marco Casadio, Colin Kessler, Rosemary Monahan, Ekaterina Komendantskaya
Abstract: Neural networks have been shown to frequently fail to learn critical safety and correctness properties purely from data, highlighting the need for training methods that directly integrate logical specifications. While adversarial training can be used to improve robustness to small perturbations within $\epsilon$-cubes, domains other than computer vision -- such as control systems and natural language processing -- may require more flexible input region specifications via generalised hyper-rectangles. Differentiable logics offer a way to encode arbitrary logical constraints as additional loss terms that guide the learning process towards satisfying these constraints. In this paper, we investigate how these two complementary approaches can be unified within a single framework for property-driven machine learning, as a step toward effective formal verification of neural networks. We show that well-known properties from the literature are subcases of this general approach, and we demonstrate its practical effectiveness on a case study involving a neural network controller for a drone system. Our framework is made publicly available at https://github.com/tflinkow/property-driven-ml.
Authors: Vincenzo De Paola, Riccardo Zamboni, Mirco Mutti, Marcello Restelli
Abstract: Parallel data collection has redefined Reinforcement Learning (RL), unlocking unprecedented efficiency and powering breakthroughs in large-scale real-world applications. In this paradigm, $N$ identical agents operate in $N$ replicas of an environment simulator, accelerating data collection by a factor of $N$. A critical question arises: \textit{Does specializing the policies of the parallel agents hold the key to surpass the $N$ factor acceleration?} In this paper, we introduce a novel learning framework that maximizes the entropy of collected data in a parallel setting. Our approach carefully balances the entropy of individual agents with inter-agent diversity, effectively minimizing redundancies. The latter idea is implemented with a centralized policy gradient method, which shows promise when evaluated empirically against systems of identical agents, as well as synergy with batch RL techniques that can exploit data diversity. Finally, we provide an original concentration analysis that shows faster rates for specialized parallel sampling distributions, which supports our methodology and may be of independent interest.
Authors: Michail Theologitis, Vasilis Samoladas, Antonios Deligiannakis
Abstract: Federated Learning (FL) enables the utilization of vast, previously inaccessible data sources. At the same time, pre-trained Language Models (LMs) have taken the world by storm and for good reason. They exhibit remarkable emergent abilities and are readily adapted to downstream tasks. This opens one of the most exciting frontiers in FL: fine-tuning LMs. Yet, a persistent challenge in FL is the frequent, rigid communication of parameters -- a problem magnified by the sheer size of these contemporary models. The FedOpt family of algorithms has become the go-to approach for FL, relying on fixed but arbitrary intervals for model exchanges. Recently, the FDA algorithm prescribed a dynamic approach by monitoring the training progress. However, it introduced a hard-to-calibrate parameter and imposed a rigid synchronization scheme. In this work, we address these limitations by proposing the FDA-Opt family of algorithms -- a unified generalization of both FDA and FedOpt. Our experimental evaluation focuses on fine-tuning LMs on downstream NLP tasks and demonstrates that FDA-Opt outperforms FedOpt even when it is configured with hyper-parameters specifically optimized for the latter. In other words, we show that FDA-Opt is a practical, drop-in replacement for FedOpt in modern FL libraries and systems: it requires no additional configuration and delivers superior performance out of the box.
Authors: William Merrill, Shane Arora, Dirk Groeneveld, Hannaneh Hajishirzi
Abstract: The right batch size is important when training language models at scale: a large batch size is necessary for fast training, but a batch size that is too large will harm token efficiency. To navigate this tradeoff, McCandlish et al. (2018) suggest that a critical batch size (CBS), below which training will not substantially degrade loss, can be estimated based on the gradient noise scale during training. While their method has been adopted in practice, e.g., when training GPT-3, strong assumptions are required to justify gradient noise as a proxy for the CBS, which makes it unclear whether their approach should be trusted in practice, limiting its applicability. In this paper, we introduce a simple, empirical approach to directly measure the CBS and show how the CBS evolves over training. Applying our approach to the OLMo models, we find that CBS is near 0 at initialization, increases rapidly at first, and then plateaus as training progresses. Furthermore, we find that this trend holds across different model sizes (1B and 7B), suggesting CBS from small training runs can inform larger-scale training runs. Our findings about how the CBS changes over training motivate batch size warmup as a natural way to reliably train language models at large batch size: start the batch size small and increase it as the CBS grows. To validate this claim, we use batch size warmup to train OLMo 1B to slightly better loss than the original training run with 43% fewer gradient steps. This shows how our framework can be applied to reliably train language models at larger batch sizes, increasing data parallelism without compromising performance.
Authors: Fu Luo, Yaoxin Wu, Zhi Zheng, Zhenkun Wang
Abstract: Recent neural combinatorial optimization (NCO) methods have shown promising problem-solving ability without requiring domain-specific expertise. Most existing NCO methods use training and testing data with a fixed constraint value and lack research on the effect of constraint tightness on the performance of NCO methods. This paper takes the capacity-constrained vehicle routing problem (CVRP) as an example to empirically analyze the NCO performance under different tightness degrees of the capacity constraint. Our analysis reveals that existing NCO methods overfit the capacity constraint, and they can only perform satisfactorily on a small range of the constraint values but poorly on other values. To tackle this drawback of existing NCO methods, we develop an efficient training scheme that explicitly considers varying degrees of constraint tightness and proposes a multi-expert module to learn a generally adaptable solving strategy. Experimental results show that the proposed method can effectively overcome the overfitting issue, demonstrating superior performances on the CVRP and CVRP with time windows (CVRPTW) with various constraint tightness degrees.
Authors: Shiangyi Lin, Hui Lan, Vasilis Syrgkanis
Abstract: Traditional instrumental variable (IV) estimators face a fundamental constraint: they can only accommodate as many endogenous treatment variables as available instruments. This limitation becomes particularly challenging in settings where the treatment is presented in a high-dimensional and unstructured manner (e.g. descriptions of patient treatment pathways in a hospital). In such settings, researchers typically resort to applying unsupervised dimension reduction techniques to learn a low-dimensional treatment representation prior to implementing IV regression analysis. We show that such methods can suffer from substantial omitted variable bias due to implicit regularization in the representation learning step. We propose a novel approach to construct treatment representations by explicitly incorporating instrumental variables during the representation learning process. Our approach provides a framework for handling high-dimensional endogenous variables with limited instruments. We demonstrate both theoretically and empirically that fitting IV models on these instrument-informed representations ensures identification of directions that optimize outcome prediction. Our experiments show that our proposed methodology improves upon the conventional two-stage approaches that perform dimension reduction without incorporating instrument information.
Authors: Ching Chang, Jeehyun Hwang, Yidan Shi, Haixin Wang, Wen-Chih Peng, Tien-Fu Chen, Wei Wang
Abstract: Time series data in real-world applications such as healthcare, climate modeling, and finance are often irregular, multimodal, and messy, with varying sampling rates, asynchronous modalities, and pervasive missingness. However, existing benchmarks typically assume clean, regularly sampled, unimodal data, creating a significant gap between research and real-world deployment. We introduce Time-IMM, a dataset specifically designed to capture cause-driven irregularity in multimodal multivariate time series. Time-IMM represents nine distinct types of time series irregularity, categorized into trigger-based, constraint-based, and artifact-based mechanisms. Complementing the dataset, we introduce IMM-TSF, a benchmark library for forecasting on irregular multimodal time series, enabling asynchronous integration and realistic evaluation. IMM-TSF includes specialized fusion modules, including a timestamp-to-text fusion module and a multimodality fusion module, which support both recency-aware averaging and attention-based integration strategies. Empirical results demonstrate that explicitly modeling multimodality on irregular time series data leads to substantial gains in forecasting performance. Time-IMM and IMM-TSF provide a foundation for advancing time series analysis under real-world conditions. The dataset is publicly available at https://www.kaggle.com/datasets/blacksnail789521/time-imm/data, and the benchmark library can be accessed at https://anonymous.4open.science/r/IMMTSF_NeurIPS2025.
URLs: https://www.kaggle.com/datasets/blacksnail789521/time-imm/data,, https://anonymous.4open.science/r/IMMTSF_NeurIPS2025.
Authors: Constantin Venhoff, Iv\'an Arcuschin, Philip Torr, Arthur Conmy, Neel Nanda
Abstract: Recent advances in large language models (LLMs) have led to the development of thinking language models that generate extensive internal reasoning chains before producing responses. While these models achieve improved performance, controlling their reasoning processes remains challenging. This work presents a steering approach for thinking LLMs by analyzing and manipulating specific reasoning behaviors in DeepSeek-R1-Distill models. Through a systematic experiment on 500 tasks across 10 diverse categories, we identify several reasoning behaviors exhibited by thinking models, including expressing uncertainty, generating examples for hypothesis validation, and backtracking in reasoning chains. We demonstrate that these behaviors are mediated by linear directions in the model's activation space and can be controlled using steering vectors. By extracting and applying these vectors, we provide a method to modulate specific aspects of the model's reasoning process, such as its tendency to backtrack or express uncertainty. Our approach offers practical tools for steering reasoning processes in thinking models in a controlled and interpretable manner. We validate our steering method using three DeepSeek-R1-Distill models, demonstrating consistent control across different model architectures.
Authors: Chenxing Wei, Jiarui Yu, Ying Tiffany He, Hande Dong, Yao Shu, Fei Yu
Abstract: DeepSeek-R1 has successfully enhanced Large Language Model (LLM) reasoning capabilities through its rule-based reward system. While it's a ''perfect'' reward system that effectively mitigates reward hacking, such reward functions are often discrete. Our experimental observations suggest that discrete rewards can lead to gradient anomaly, unstable optimization, and slow convergence. To address this issue, we propose ReDit (Reward Dithering), a method that dithers the discrete reward signal by adding simple random noise. With this perturbed reward, exploratory gradients are continuously provided throughout the learning process, enabling smoother gradient updates and accelerating convergence. The injected noise also introduces stochasticity into flat reward regions, encouraging the model to explore novel policies and escape local optima. Experiments across diverse tasks demonstrate the effectiveness and efficiency of ReDit. On average, ReDit achieves performance comparable to vanilla GRPO with only approximately 10% the training steps, and furthermore, still exhibits a 4% performance improvement over vanilla GRPO when trained for a similar duration. Visualizations confirm significant mitigation of gradient issues with ReDit. Moreover, theoretical analyses are provided to further validate these advantages.
Authors: Huiling Zhang, Zi Xu, Yu-Hong Dai
Abstract: The study of nonconvex minimax games has gained significant momentum in machine learning and decision science communities due to their fundamental connections to adversarial training scenarios. This work develops a primal-dual alternating proximal gradient (PDAPG) algorithm framework for resolving iterative minimax games featuring nonsmooth nonconvex objectives subject to coupled linear constraints. We establish rigorous convergence guarantees for both nonconvex-strongly concave and nonconvex-concave game configurations, demonstrating that PDAPG achieves an $\varepsilon$-stationary solution within $\mathcal{O}\left( \varepsilon ^{-2} \right)$ iterations for strongly concave settings and $\mathcal{O}\left( \varepsilon ^{-4} \right)$ iterations for concave scenarios. Our analysis provides the first known iteration complexity bounds for this class of constrained minimax games, particularly addressing the critical challenge of coupled linear constraints that induce inherent interdependencies among strategy variables. The proposed game-theoretic framework advances existing solution methodologies by simultaneously handling nonsmooth components and coordinated constraint structures through alternating primal-dual updates.
Authors: Julia Ackermann, Arnulf Jentzen, Thomas Kruse, Benno Kuckuck, Joshua Lee Padgett
Abstract: Recently, several deep learning (DL) methods for approximating high-dimensional partial differential equations (PDEs) have been proposed. The interest that these methods have generated in the literature is in large part due to simulations which appear to demonstrate that such DL methods have the capacity to overcome the curse of dimensionality (COD) for PDEs in the sense that the number of computational operations they require to achieve a certain approximation accuracy $\varepsilon\in(0,\infty)$ grows at most polynomially in the PDE dimension $d\in\mathbb N$ and the reciprocal of $\varepsilon$. While there is thus far no mathematical result that proves that one of such methods is indeed capable of overcoming the COD, there are now a number of rigorous results in the literature that show that deep neural networks (DNNs) have the expressive power to approximate PDE solutions without the COD in the sense that the number of parameters used to describe the approximating DNN grows at most polynomially in both the PDE dimension $d\in\mathbb N$ and the reciprocal of the approximation accuracy $\varepsilon>0$. Roughly speaking, in the literature it is has been proved for every $T>0$ that solutions $u_d\colon [0,T]\times\mathbb R^d\to \mathbb R$, $d\in\mathbb N$, of semilinear heat PDEs with Lipschitz continuous nonlinearities can be approximated by DNNs with ReLU activation at the terminal time in the $L^2$-sense without the COD provided that the initial value functions $\mathbb R^d\ni x\mapsto u_d(0,x)\in\mathbb R$, $d\in\mathbb N$, can be approximated by ReLU DNNs without the COD. It is the key contribution of this work to generalize this result by establishing this statement in the $L^p$-sense with $p\in(0,\infty)$ and by allowing the activation function to be more general covering the ReLU, the leaky ReLU, and the softplus activation functions as special cases.
Authors: Shafagh Keyvanian, Michelle J. Johnson, Nadia Figueroa
Abstract: A realistic human kinematic model that satisfies anatomical constraints is essential for human-robot interaction, biomechanics and robot-assisted rehabilitation. Modeling realistic joint constraints, however, is challenging as human arm motion is constrained by joint limits, inter- and intra-joint dependencies, self-collisions, individual capabilities and muscular or neurological constraints which are difficult to represent. Hence, physicians and researchers have relied on simple box-constraints, ignoring important anatomical factors. In this paper, we propose a data-driven method to learn realistic anatomically constrained upper-limb range of motion (RoM) boundaries from motion capture data. This is achieved by fitting a one-class support vector machine to a dataset of upper-limb joint space exploration motions with an efficient hyper-parameter tuning scheme. Our approach outperforms similar works focused on valid RoM learning. Further, we propose an impairment index (II) metric that offers a quantitative assessment of capability/impairment when comparing healthy and impaired arms. We validate the metric on healthy subjects physically constrained to emulate hemiplegia and different disability levels as stroke patients. [https://sites.google.com/seas.upenn.edu/learning-rom]
Authors: Justin Kay, Timm Haucke, Suzanne Stathatos, Siqi Deng, Erik Young, Pietro Perona, Sara Beery, Grant Van Horn
Abstract: Object detectors often perform poorly on data that differs from their training set. Domain adaptive object detection (DAOD) methods have recently demonstrated strong results on addressing this challenge. Unfortunately, we identify systemic benchmarking pitfalls that call past results into question and hamper further progress: (a) Overestimation of performance due to underpowered baselines, (b) Inconsistent implementation practices preventing transparent comparisons of methods, and (c) Lack of generality due to outdated backbones and lack of diversity in benchmarks. We address these problems by introducing: (1) A unified benchmarking and implementation framework, Align and Distill (ALDI), enabling comparison of DAOD methods and supporting future development, (2) A fair and modern training and evaluation protocol for DAOD that addresses benchmarking pitfalls, (3) A new DAOD benchmark dataset, CFC-DAOD, enabling evaluation on diverse real-world data, and (4) A new method, ALDI++, that achieves state-of-the-art results by a large margin. ALDI++ outperforms the previous state-of-the-art by +3.5 AP50 on Cityscapes to Foggy Cityscapes, +5.7 AP50 on Sim10k to Cityscapes (where ours is the only method to outperform a fair baseline), and +0.6 AP50 on CFC Kenai to Channel. ALDI and ALDI++ are architecture-agnostic, setting a new state-of-the-art for YOLO and DETR-based DAOD as well without additional hyperparameter tuning. Our framework, dataset, and state-of-the-art method offer a critical reset for DAOD and provide a strong foundation for future research. Code and data are available: https://github.com/justinkay/aldi and https://github.com/visipedia/caltech-fish-counting.
URLs: https://github.com/justinkay/aldi, https://github.com/visipedia/caltech-fish-counting.
Authors: Sophie Starck, Vasiliki Sideri-Lampretsa, Bernhard Kainz, Martin J. Menten, Tamara T. Mueller, Daniel Rueckert
Abstract: Anatomical atlases are widely used for population studies and analysis. Conditional atlases target a specific sub-population defined via certain conditions, such as demographics or pathologies, and allow for the investigation of fine-grained anatomical differences like morphological changes associated with ageing or disease. Existing approaches use either registration-based methods that are often unable to handle large anatomical variations or generative adversarial models, which are challenging to train since they can suffer from training instabilities. Instead of generating atlases directly in as intensities, we propose using latent diffusion models to generate deformation fields, which transform a general population atlas into one representing a specific sub-population. Our approach ensures structural integrity, enhances interpretability and avoids hallucinations that may arise during direct image synthesis by generating this deformation field and regularising it using a neighbourhood of images. We compare our method to several state-of-the-art atlas generation methods using brain MR images from the UK Biobank. Our method generates highly realistic atlases with smooth transformations and high anatomical fidelity, outperforming existing baselines. We demonstrate the quality of these atlases through comprehensive evaluations, including quantitative metrics for anatomical accuracy, perceptual similarity, and qualitative analyses displaying the consistency and realism of the generated atlases.
Authors: Ting Luo, Jing Zhang, Yingwei Qiu, Li Zhang, Yaohua Hu, Zhuliang Yu, Zhen Liang
Abstract: Emotion decoding using Electroencephalography (EEG)-based affective brain-computer interfaces (aBCIs) plays a crucial role in affective computing but is limited by challenges such as EEG's non-stationarity, individual variability, and the high cost of large labeled datasets. While deep learning methods are effective, they require extensive computational resources and large data volumes, limiting their practical application. To overcome these issues, we propose Manifold-based Domain Adaptation with Dynamic Distribution (M3D), a lightweight, non-deep transfer learning framework. M3D consists of four key modules: manifold feature transformation, dynamic distribution alignment, classifier learning, and ensemble learning. The data is mapped to an optimal Grassmann manifold space, enabling dynamic alignment of source and target domains. This alignment is designed to prioritize both marginal and conditional distributions, improving adaptation efficiency across diverse datasets. In classifier learning, the principle of structural risk minimization is applied to build robust classification models. Additionally, dynamic distribution alignment iteratively refines the classifier. The ensemble learning module aggregates classifiers from different optimization stages to leverage diversity and enhance prediction accuracy. M3D is evaluated on two EEG emotion recognition datasets using two validation protocols (cross-subject single-session and cross-subject cross-session) and a clinical EEG dataset for Major Depressive Disorder (MDD). Experimental results show that M3D outperforms traditional non-deep learning methods with a 4.47% average improvement and achieves deep learning-level performance with reduced data and computational requirements, demonstrating its potential for real-world aBCI applications.
Authors: Shadab Ahamed, Arman Rahmim
Abstract: Minimizing the need for pixel-level annotated data to train PET lesion detection and segmentation networks is highly desired and can be transformative, given time and cost constraints associated with expert annotations. Current unsupervised or weakly-supervised anomaly detection methods rely on autoencoder or generative adversarial networks (GANs) trained only on healthy data. While these approaches reduce annotation dependency, GAN-based methods are notably more challenging to train than non-GAN alternatives (such as autoencoders) due to issues such as the simultaneous optimization of two competing networks, mode collapse, and training instability. In this paper, we present the weakly-supervised $\textbf{I}$mplicitly-$\textbf{g}$uided $\textbf{CO}$u$\textbf{N}$terfactual diffusion model for $\textbf{D}$etecting $\textbf{A}$nomalies in $\textbf{PET}$ images (IgCONDA-PET). The solution is developed and validated using PET scans from six retrospective cohorts consisting of a total of 2652 cases (multi-cancer, multi-tracer) containing both local and public datasets (spanning multiple centers). The training is conditioned on image class labels (healthy vs. unhealthy) via attention modules, and we employ implicit diffusion guidance. We perform counterfactual generation which facilitates "unhealthy-to-healthy" domain translation by generating a synthetic, healthy version of an unhealthy input image, enabling the detection of anomalies through the calculated differences. The performance of our method was compared against several other deep learning based weakly-supervised or unsupervised methods as well as traditional methods like 41% SUV$_\text{max}$ thresholding. We also highlight the importance of incorporating attention modules in our network for the detection of small anomalies. The code is publicly available at: https://github.com/ahxmeds/IgCONDA-PET.git.
Authors: Saurabh Saini, Kapil Ahuja, Akshat S. Chauhan
Abstract: Cervical cancer is the second most common cancer in women and causes high death rates. Earlier models for detecting cervical cancer had limited success. In this work, we propose new models that substantially outperform previous models. Previous studies show that pretrained ResNets extract features from cervical cancer images well. Hence, our first model involves working with three ResNets (50, 101, 152). All the existing works use only the last convolution block of their respective ResNet, which captures abstract features (e.g., shapes, objects). However, we believe that detailed features (e.g., color, edges, texture), coming from earlier convolution blocks, are equally important for cancer (specifically cervical cancer) classification. Since now the number of features become large, we use a novel feature selection technique of Global Max Pooling for detailed features and Global Average Pooling for abstract features. Hence, our second model consists of the resulting Cascaded Block Fused variants of the three ResNets. To improve the performance further, we combine and normalize the features of the three standard ResNets as well as our proposed three Cascaded Block Fused ResNets. This type of combination is also new in cancer classification domain (also in cervical cancer), and results in our third and fourth models, respectively. We use a linear SVM for classification. We exhaustively perform experiments on two public datasets, IARC and AnnoCerv, achieving an average performance of 97.92% and 92.97% surpassing standard ResNets performance of 90.89% and 87.97%, respectively. We outperform the competitive approach available on IARC dataset with an average gain of 13.20%, while no prior competitive work available on AnnoCerv. Additionally, we introduce a novel SHAP+LIME explainability method, accurately identifying the cancerous region in 97% of cases.
Authors: Nathan T. Riek, Murat Akcakaya, Zeineb Bouzid, Tanmay Gokhale, Stephanie Helman, Karina Kraevsky-Philips, Rui Qi Ji, Ervin Sejdic, Jessica K. Z\`egre-Hemsey, Christian Martin-Gill, Clifton W. Callaway, Samir Saba, Salah Al-Zaiti
Abstract: Objective: In this paper we develop and evaluate ECG-SMART-NET for occlusion myocardial infarction (OMI) identification. OMI is a severe form of heart attack characterized by complete blockage of one or more coronary arteries requiring immediate referral for cardiac catheterization to restore blood flow to the heart. Two thirds of OMI cases are difficult to visually identify from a 12-lead electrocardiogram (ECG) and can be potentially fatal if not identified quickly. Previous works on this topic are scarce, and current state-of-the-art evidence suggests both feature-based random forests and convolutional neural networks (CNNs) are promising approaches to improve ECG detection of OMI. Methods: While the ResNet architecture has been adapted for use with ECG recordings, it is not ideally suited to capture informative temporal features within each lead and the spatial concordance or discordance across leads. We propose a clinically informed modification of the ResNet-18 architecture. The model first learns temporal features through temporal convolutional layers with 1xk kernels followed by a spatial convolutional layer, after the residual blocks, with 12x1 kernels to learn spatial features. Results: ECG-SMART-NET was benchmarked against the original ResNet-18 and other state-of-the-art models on a multisite real-word clinical dataset that consists of 10,393 ECGs from 7,397 unique patients (rate of OMI =7.2%). ECG-SMART-NET outperformed other models in the classification of OMI with a test AUC of 0.953 [0.921, 0.978]. Conclusion and Significance: ECG-SMART-NET can outperform the state-of-the-art random forest for OMI prediction and is better suited for this task than the original ResNet-18 architecture.
Authors: Mart\'in Hern\'andez, Enrique Zuazua
Abstract: We present a fully constructive analysis of deep ReLU neural networks for classification and function approximation tasks. First, we prove that any dataset with $N$ distinct points in $\mathbb{R}^d$ and $M$ output classes can be exactly classified using a multilayer perceptron (MLP) of width $2$ and depth at most $2N + 4M - 1$, with all network parameters constructed explicitly. This result is sharp with respect to width and is interpreted through the lens of simultaneous or ensemble controllability in discrete nonlinear dynamics. Second, we show that these explicit constructions yield uniform bounds on the parameter norms and, in particular, provide upper estimates for minimizers of standard regularized training loss functionals in supervised learning. As the regularization parameter vanishes, the trained networks converge to exact classifiers with bounded norm, explaining the effectiveness of overparameterized training in the small-regularization regime. We also prove a universal approximation theorem in $L^p(\Omega; \mathbb{R}_+)$ for any bounded domain $\Omega \subset \mathbb{R}^d$ and $p \in [1, \infty)$, using MLPs of fixed width $d + 1$. The proof is constructive, geometrically motivated, and provides explicit estimates on the network depth when the target function belongs to the Sobolev space $W^{1,p}$. We also extend the approximation and depth estimation results to $L^p(\Omega; \mathbb{R}^m)$ for any $m \geq 1$. Our results offer a unified and interpretable framework connecting controllability, expressivity, and training dynamics in deep neural networks.
Authors: C. Nicol\`o De Sabbata, Theodore R. Sumers, Badr AlKhamissi, Antoine Bosselut, Thomas L. Griffiths
Abstract: Being prompted to engage in reasoning has emerged as a core technique for using large language models (LLMs), deploying additional inference-time compute to improve task performance. However, as LLMs increase in both size and adoption, inference costs are correspondingly becoming increasingly burdensome. How, then, might we optimize reasoning's cost-performance tradeoff? This work introduces a novel approach based on computational models of metareasoning used in cognitive science, training LLMs to selectively use intermediate reasoning steps only when necessary. We first develop a reward function that incorporates the Value of Computation by penalizing unnecessary reasoning, then use this reward function with Expert Iteration to train the LLM. Compared to few-shot chain-of-thought prompting and STaR, our method significantly reduces inference costs (20-37\% fewer tokens generated across three models) while maintaining task performance across diverse datasets.
Authors: Chung-En Sun, Xiaodong Liu, Weiwei Yang, Tsui-Wei Weng, Hao Cheng, Aidan San, Michel Galley, Jianfeng Gao
Abstract: Recent research has shown that Large Language Models (LLMs) are vulnerable to automated jailbreak attacks, where adversarial suffixes crafted by algorithms appended to harmful queries bypass safety alignment and trigger unintended responses. Current methods for generating these suffixes are computationally expensive and have low Attack Success Rates (ASR), especially against well-aligned models like Llama2 and Llama3. To overcome these limitations, we introduce ADV-LLM, an iterative self-tuning process that crafts adversarial LLMs with enhanced jailbreak ability. Our framework significantly reduces the computational cost of generating adversarial suffixes while achieving nearly 100\% ASR on various open-source LLMs. Moreover, it exhibits strong attack transferability to closed-source models, achieving 99\% ASR on GPT-3.5 and 49\% ASR on GPT-4, despite being optimized solely on Llama3. Beyond improving jailbreak ability, ADV-LLM provides valuable insights for future safety alignment research through its ability to generate large datasets for studying LLM safety.
Authors: Azim Ospanov, Farzan Farnia
Abstract: Evaluating the diversity of generative models without reference data poses methodological challenges. The reference-free Vendi and RKE scores address this by quantifying the diversity of generated data using matrix-based entropy measures. Among these two, the Vendi score is typically computed via the eigendecomposition of an $n \times n$ kernel matrix constructed from n generated samples. However, the prohibitive computational cost of eigendecomposition for large $n$ often limits the number of samples used to fewer than 20,000. In this paper, we investigate the statistical convergence of the Vendi and RKE scores under restricted sample sizes. We numerically demonstrate that, in general, the Vendi score computed with standard sample sizes below 20,000 may not converge to its asymptotic value under infinite sampling. To address this, we introduce the $t$-truncated Vendi score by truncating the eigenspectrum of the kernel matrix, which is provably guaranteed to converge to its population limit with $n=\mathcal{O}(t)$ samples. We further show that existing Nystr\"om and FKEA approximation methods converge to the asymptotic limit of the truncated Vendi score. In contrast to the Vendi score, we prove that the RKE score enjoys universal convergence guarantees across all kernel functions. We conduct several numerical experiments to illustrate the concentration of Nystr\"om and FKEA computed Vendi scores around the truncated Vendi score, and we analyze how the truncated Vendi and RKE scores correlate with the diversity of image and text data. The code is available at https://github.com/aziksh-ospanov/truncated-vendi.
Authors: Anish Dhir, Ruby Sedgwick, Avinash Kori, Ben Glocker, Mark van der Wilk
Abstract: Current causal discovery approaches require restrictive model assumptions in the absence of interventional data to ensure structure identifiability. These assumptions often do not hold in real-world applications leading to a loss of guarantees and poor performance in practice. Recent work has shown that, in the bivariate case, Bayesian model selection can greatly improve performance by exchanging restrictive modelling for more flexible assumptions, at the cost of a small probability of making an error. Our work shows that this approach is useful in the important multivariate case as well. We propose a scalable algorithm leveraging a continuous relaxation of the discrete model selection problem. Specifically, we employ the Causal Gaussian Process Conditional Density Estimator (CGP-CDE) as a Bayesian non-parametric model, using its hyperparameters to construct an adjacency matrix. This matrix is then optimised using the marginal likelihood and an acyclicity regulariser, giving the maximum a posteriori causal graph. We demonstrate the competitiveness of our approach, showing it is advantageous to perform multivariate causal discovery without infeasible assumptions using Bayesian model selection.
Authors: Julian Junyan Wang, Victor Xiaoqi Wang
Abstract: Unequal access to costly datasets essential for empirical research has long hindered researchers from disadvantaged institutions, limiting their ability to contribute to their fields and advance their careers. Recent breakthroughs in Large Language Models (LLMs) have the potential to democratize data access by automating data collection from unstructured sources. We develop and evaluate a novel methodology using GPT-4o-mini within a Retrieval-Augmented Generation (RAG) framework to collect data from corporate disclosures. Our approach achieves human-level accuracy in collecting CEO pay ratios from approximately 10,000 proxy statements and Critical Audit Matters (CAMs) from more than 12,000 10-K filings, with LLM processing times of 9 and 40 minutes respectively, each at a cost under $10. This stands in stark contrast to the hundreds of hours needed for manual collection or the thousands of dollars required for commercial database subscriptions. To foster a more inclusive research community by empowering researchers with limited resources to explore new avenues of inquiry, we share our methodology and the resulting datasets.
Authors: Aur\'elien Decelle, Alfonso de Jes\'us Navas G\'omez, Beatriz Seoane
Abstract: Maximum entropy methods, rooted in the inverse Ising/Potts problem from statistical physics, are widely used to model pairwise interactions in complex systems across disciplines such as bioinformatics and neuroscience. While successful, these approaches often fail to capture higher-order interactions that are critical for understanding collective behavior. In contrast, modern machine learning methods can model such interactions, but their interpretability often comes at a prohibitive computational cost. Restricted Boltzmann Machines (RBMs) provide a computationally efficient alternative by encoding statistical correlations through hidden units in a bipartite architecture. In this work, we introduce a method that maps RBMs onto generalized Potts models, enabling the systematic extraction of interactions up to arbitrary order. Leveraging large-$N$ approximations -- made tractable by the RBM's structure -- we extract effective many-body couplings with minimal computational effort. We further propose a robust framework for recovering higher-order interactions in more complex generative models, and introduce a simple gauge-fixing scheme for the effective Potts representation. Validation on synthetic data demonstrates accurate recovery of two- and three-body interactions. Applied to protein sequence data, our method reconstructs contact maps with high fidelity and outperforms state-of-the-art inverse Potts models. These results establish RBMs as a powerful and efficient tool for modeling higher-order structure in high-dimensional categorical data.
Authors: Emad Efatinasab, Alessandro Brighente, Denis Donadel, Mauro Conti, Mirco Rampazzo
Abstract: Smart grids are crucial for meeting rising energy demands driven by global population growth and urbanization. By integrating renewable energy sources, they enhance efficiency, reliability, and sustainability. However, ensuring their availability and security requires advanced operational control and safety measures. Although artificial intelligence and machine learning can help assess grid stability, challenges such as data scarcity and cybersecurity threats, particularly adversarial attacks, remain. Data scarcity is a major issue, as obtaining real-world instances of grid instability requires significant expertise, resources, and time. Yet, these instances are critical for testing new research advancements and security mitigations. This paper introduces a novel framework for detecting instability in smart grids using only stable data. It employs a Generative Adversarial Network (GAN) where the generator is designed not to produce near-realistic data but instead to generate Out-Of-Distribution (OOD) samples with respect to the stable class. These OOD samples represent unstable behavior, anomalies, or disturbances that deviate from the stable data distribution. By training exclusively on stable data and exposing the discriminator to OOD samples, our framework learns a robust decision boundary to distinguish stable conditions from any unstable behavior, without requiring unstable data during training. Furthermore, we incorporate an adversarial training layer to enhance resilience against attacks. Evaluated on a real-world dataset, our solution achieves up to 98.1\% accuracy in predicting grid stability and 98.9\% in detecting adversarial attacks. Implemented on a single-board computer, it enables real-time decision-making with an average response time of under 7ms.
Authors: Galen Reeves, Henry D. Pfister
Abstract: This paper provides an elementary, self-contained analysis of diffusion-based sampling methods for generative modeling. In contrast to existing approaches that rely on continuous-time processes and then discretize, our treatment works directly with discrete-time stochastic processes and yields precise non-asymptotic convergence guarantees under broad assumptions. The key insight is to couple the sampling process of interest with an idealized comparison process that has an explicit Gaussian-convolution structure. We then leverage simple identities from information theory, including the I-MMSE relationship, to bound the discrepancy (in terms of the Kullback-Leibler divergence) between these two discrete-time processes. In particular, we show that, if the diffusion step sizes are chosen sufficiently small and one can approximate certain conditional mean estimators well, then the sampling distribution is provably close to the target distribution. Our results also provide a transparent view on how to accelerate convergence by using additional randomness in each step to match higher-order moments in the comparison process.
Authors: Antoni Kowalczuk, Jan Dubi\'nski, Franziska Boenisch, Adam Dziedzic
Abstract: Image AutoRegressive generation has emerged as a new powerful paradigm with image autoregressive models (IARs) matching state-of-the-art diffusion models (DMs) in image quality (FID: 1.48 vs. 1.58) while allowing for a higher generation speed. However, the privacy risks associated with IARs remain unexplored, raising concerns regarding their responsible deployment. To address this gap, we conduct a comprehensive privacy analysis of IARs, comparing their privacy risks to the ones of DMs as reference points. Concretely, we develop a novel membership inference attack (MIA) that achieves a remarkably high success rate in detecting training images (with a True Positive Rate at False Positive Rate = 1% of 86.38% vs. 6.38% for DMs with comparable attacks). We leverage our novel MIA to provide dataset inference (DI) for IARs, and show that it requires as few as 6 samples to detect dataset membership (compared to 200 for DI in DMs), confirming a higher information leakage in IARs. Finally, we are able to extract hundreds of training data points from an IAR (e.g., 698 from VAR-d30). Our results suggest a fundamental privacy-utility trade-off: while IARs excel in image generation quality and speed, they are empirically significantly more vulnerable to privacy attacks compared to DMs that achieve similar performance. We release the code at https://github.com/sprintml/privacy_attacks_against_iars for reproducibility.
URLs: https://github.com/sprintml/privacy_attacks_against_iars
Authors: Yifan Shen, Peiyuan Zhu, Zijian Li, Shaoan Xie, Zeyu Tang, Namrata Deka, Zongfang Liu, Guangyi Chen, Kun Zhang
Abstract: Controllable video generation remains a significant challenge, despite recent advances in generating high-quality and consistent videos. Most existing methods for controlling video generation treat the video as a whole, neglecting intricate fine-grained spatiotemporal relationships, which limits both control precision and efficiency. In this paper, we propose Controllable Video Generative Adversarial Networks (CoVoGAN) to disentangle the video concepts, thus facilitating efficient and independent control over individual concepts. Specifically, following the minimal change principle, we first disentangle static and dynamic latent variables. We then leverage the sufficient change property to achieve component-wise identifiability of dynamic latent variables, enabling disentangled control of video generation. To establish the theoretical foundation, we provide a rigorous analysis demonstrating the identifiability of our approach. Building on these theoretical insights, we design a Temporal Transition Module to disentangle latent dynamics. To enforce the minimal change principle and sufficient change property, we minimize the dimensionality of latent dynamic variables and impose temporal conditional independence. To validate our approach, we integrate this module as a plug-in for GANs. Extensive qualitative and quantitative experiments on various video generation benchmarks demonstrate that our method significantly improves generation quality and controllability across diverse real-world scenarios.
Authors: Georg B\"okman, David Nordstr\"om, Fredrik Kahl
Abstract: Incorporating geometric invariance into neural networks enhances parameter efficiency but typically increases computational costs. This paper introduces new equivariant neural networks that preserve symmetry while maintaining a comparable number of floating-point operations (FLOPs) per parameter to standard non-equivariant networks. We focus on horizontal mirroring (flopping) invariance, common in many computer vision tasks. The main idea is to parametrize the feature spaces in terms of mirror-symmetric and mirror-antisymmetric features, i.e., irreps of the flopping group. This decomposes the linear layers to be block-diagonal, requiring half the number of FLOPs. Our approach reduces both FLOPs and wall-clock time, providing a practical solution for efficient, scalable symmetry-aware architectures.
Authors: Yuan Sui, Yufei He, Tri Cao, Simeng Han, Yulin Chen, Bryan Hooi
Abstract: Large Language Models (LLMs) increasingly rely on prolonged reasoning chains to solve complex tasks. However, this trial-and-error approach often leads to high computational overhead and error propagation, where early mistakes can derail subsequent steps. To address these issues, we introduce Meta-Reasoner, a framework that dynamically optimizes inference-time reasoning by enabling LLMs to "think about how to think." Drawing inspiration from human meta-cognition and dual-process theory, Meta-Reasoner operates as a strategic advisor, decoupling high-level guidance from step-by-step generation. It employs contextual multi-armed bandits to iteratively evaluate reasoning progress and select optimal strategies (e.g., backtrack, clarify ambiguity, restart from scratch, or propose alternative approaches), and reallocates computational resources toward the most promising paths. Our evaluations on mathematical reasoning and puzzles highlight the potential of dynamic reasoning chains to overcome inherent challenges in the LLM reasoning process and also show promise in broader applications, offering a scalable and adaptable solution for reasoning-intensive tasks.
Authors: Sara Rajaee, Kumar Pratik, Gabriele Cesa, Arash Behboodi
Abstract: The most promising recent methods for AI reasoning require applying variants of reinforcement learning (RL) either on rolled out trajectories from the LLMs, even for the step-wise rewards, or large quantities of human-annotated trajectory data. The reliance on the rolled-out trajectory renders the compute cost and time prohibitively high. In particular, the correctness of a reasoning trajectory can typically only be judged at its completion, leading to sparse rewards in RL or requiring expensive synthetic data generation in expert iteration-like methods. In this work, we focus on the Automatic Theorem Proving (ATP) task and propose a novel verifier-in-the-loop design, which, unlike existing approaches that leverage feedback on the entire reasoning trajectory, employs an automated verifier to give intermediate feedback at each step of the reasoning process. Using Lean as the verifier, we empirically show that the step-by-step local verification produces a global improvement in the model's reasoning accuracy and efficiency.
Authors: Kai Ye, Hongyi Zhou, Jin Zhu, Francesco Quinzan, Chengchun Shi
Abstract: Reinforcement learning from human feedback (RLHF) has emerged as a key technique for aligning the output of large language models (LLMs) with human preferences. To learn the reward function, most existing RLHF algorithms use the Bradley-Terry model, which relies on assumptions about human preferences that may not reflect the complexity and variability of real-world judgments. In this paper, we propose a robust algorithm to enhance the performance of existing approaches under such reward model misspecifications. Theoretically, our algorithm reduces the variance of reward and policy estimators, leading to improved regret bounds. Empirical evaluations on LLM benchmark datasets demonstrate that the proposed algorithm consistently outperforms existing methods, with 77-81% of responses being favored over baselines on the Anthropic Helpful and Harmless dataset.
Authors: Hari Sarang Nathan
Abstract: The Hausdorff dimension of a set can be detected using the Riesz energy. Here, we consider situations where a sequence of points, $\{x_n\}$, ``fills in'' a set $E \subset \mathbb{R}^d$ in an appropriate sense and investigate the degree to which the discrete analog to the Riesz energy of these sets can be used to bound the Hausdorff dimension of $E$. We also discuss applications to data science and Erd\H{o}s/Falconer type problems.
Authors: Mohamed Moustafa, Joseph Lemley, Peter Corcoran
Abstract: Time event cameras are a novel technology for recording scene information at extremely low latency and with low power consumption. Event cameras output a stream of events that encapsulate pixel-level light intensity changes within the scene, capturing information with a higher dynamic range and temporal resolution than traditional cameras. This study investigates the contact-free reconstruction of an individual's cardiac pulse signal from time event recording of their face using a supervised convolutional neural network (CNN) model. An end-to-end model is trained to extract the cardiac signal from a two-dimensional representation of the event stream, with model performance evaluated based on the accuracy of the calculated heart rate. The experimental results confirm that physiological cardiac information in the facial region is effectively preserved within the event stream, showcasing the potential of this novel sensor for remote heart rate monitoring. The model trained on event frames achieves a root mean square error (RMSE) of 3.32 beats per minute (bpm) compared to the RMSE of 2.92 bpm achieved by the baseline model trained on standard camera frames. Furthermore, models trained on event frames generated at 60 and 120 FPS outperformed the 30 FPS standard camera results, achieving an RMSE of 2.54 and 2.13 bpm, respectively.
Authors: Theo Lepage, Reda Dehak
Abstract: Self-Supervised Learning (SSL) has led to considerable progress in Speaker Verification (SV). The standard framework uses same-utterance positive sampling and data-augmentation to generate anchor-positive pairs of the same speaker. This is a major limitation, as this strategy primarily encodes channel information from the recording condition, shared by the anchor and positive. We propose a new positive sampling technique to address this bottleneck: Self-Supervised Positive Sampling (SSPS). For a given anchor, SSPS aims to find an appropriate positive, i.e., of the same speaker identity but a different recording condition, in the latent space using clustering assignments and a memory queue of positive embeddings. SSPS improves SV performance for both SimCLR and DINO, reaching 2.57% and 2.53% EER, outperforming SOTA SSL methods on VoxCeleb1-O. In particular, SimCLR-SSPS achieves a 58% EER reduction by lowering intra-speaker variance, providing comparable performance to DINO-SSPS.
Authors: Hongyuan Tao, Ying Zhang, Zhenhao Tang, Hongen Peng, Xukun Zhu, Bingchang Liu, Yingguang Yang, Ziyin Zhang, Zhaogui Xu, Haipeng Zhang, Linchao Zhu, Rui Wang, Hang Yu, Jianguo Li, Peng Di
Abstract: Recent advances in Large Language Models (LLMs) have shown promise in function-level code generation, yet repository-level software engineering tasks remain challenging. Current solutions predominantly rely on proprietary LLM agents, which introduce unpredictability and limit accessibility, raising concerns about data privacy and model customization. This paper investigates whether open-source LLMs can effectively address repository-level tasks without requiring agent-based approaches. We demonstrate this is possible by enabling LLMs to comprehend functions and files within codebases through their semantic information and structural dependencies. To this end, we introduce Code Graph Models (CGMs), which integrate repository code graph structures into the LLM's attention mechanism and map node attributes to the LLM's input space using a specialized adapter. When combined with an agentless graph RAG framework, our approach achieves a 43.00% resolution rate on the SWE-bench Lite benchmark using the open-source Qwen2.5-72B model. This performance ranks first among open weight models, second among methods with open-source systems, and eighth overall, surpassing the previous best open-source model-based method by 12.33%.
Authors: Daniel Jarne Ornia, Nicholas Bishop, Joel Dyer, Wei-Chen Lee, Ani Calinescu, Doyne Farmer, Michael Wooldridge
Abstract: Advanced reasoning models with agentic capabilities (AI agents) are deployed to interact with humans and to solve sequential decision-making problems under (approximate) utility functions and internal models. When such problems have resource or failure constraints where action sequences may be forcibly terminated once resources are exhausted, agents face implicit trade-offs that reshape their utility-driven (rational) behaviour. Additionally, since these agents are typically commissioned by a human principal to act on their behalf, asymmetries in constraint exposure can give rise to previously unanticipated misalignment between human objectives and agent incentives. We formalise this setting through a survival bandit framework, provide theoretical and empirical results that quantify the impact of survival-driven preference shifts, identify conditions under which misalignment emerges and propose mechanisms to mitigate the emergence of risk-seeking or risk-averse behaviours. As a result, this work aims to increase understanding and interpretability of emergent behaviours of AI agents operating under such survival pressure, and offer guidelines for safely deploying such AI systems in critical resource-limited environments.
Authors: Alan Chen, Jack Merullo, Alessandro Stolfo, Ellie Pavlick
Abstract: In this work, we demonstrate that affine mappings between residual streams of language models is a cheap way to effectively transfer represented features between models. We apply this technique to transfer the weights of Sparse Autoencoders (SAEs) between models of different sizes to compare their representations. We find that small and large models learn similar representation spaces, which motivates training expensive components like SAEs on a smaller model and transferring to a larger model at a FLOPs savings. In particular, using a small-to-large transferred SAE as initialization can lead to 50% cheaper training runs when training SAEs on larger models. Next, we show that transferred probes and steering vectors can effectively recover ground truth performance. Finally, we dive deeper into feature-level transferability, finding that semantic and structural features transfer noticeably differently while specific classes of functional features have their roles faithfully mapped. Overall, our findings illustrate similarities and differences in the linear representation spaces of small and large models and demonstrate a method for improving the training efficiency of SAEs.
Authors: Francesco Vinci, Gyunam Park, Wil van der Aalst, Massimiliano de Leoni
Abstract: Business Process Simulation (BPS) refers to techniques designed to replicate the dynamic behavior of a business process. Many approaches have been proposed to automatically discover simulation models from historical event logs, reducing the cost and time to manually design them. However, in dynamic business environments, organizations continuously refine their processes to enhance efficiency, reduce costs, and improve customer satisfaction. Existing techniques to process simulation discovery lack adaptability to real-time operational changes. In this paper, we propose a streaming process simulation discovery technique that integrates Incremental Process Discovery with Online Machine Learning methods. This technique prioritizes recent data while preserving historical information, ensuring adaptation to evolving process dynamics. Experiments conducted on four different event logs demonstrate the importance in simulation of giving more weight to recent data while retaining historical knowledge. Our technique not only produces more stable simulations but also exhibits robustness in handling concept drift, as highlighted in one of the use cases.
Authors: Ziv Scully, Alexander Terenin
Abstract: The Gittins index is a tool that optimally solves a variety of decision-making problems involving uncertainty, including multi-armed bandit problems, minimizing mean latency in queues, and search problems like the Pandora's box model. However, despite the above examples and later extensions thereof, the space of problems that the Gittins index can solve perfectly optimally is limited, and its definition is rather subtle compared to those of other multi-armed bandit algorithms. As a result, the Gittins index is often regarded as being primarily a concept of theoretical importance, rather than a practical tool for solving decision-making problems. The aim of this tutorial is to demonstrate that the Gittins index can be fruitfully applied to practical problems. We start by giving an example-driven introduction to the Gittins index, then walk through several examples of problems it solves - some optimally, some suboptimally but still with excellent performance. Two practical highlights in the latter category are applying the Gittins index to Bayesian optimization, and applying the Gittins index to minimizing tail latency in queues.
Authors: Tawsif Ahmed, Andrej Radonjic, Gollam Rabby
Abstract: We present Sleeping-DISCO 9M, a large-scale pre-training dataset for music and song. To the best of our knowledge, there are no open-source high-quality dataset representing popular and well-known songs for generative music modeling tasks such as text-music, music-captioning, singing-voice synthesis, melody reconstruction and cross-model retrieval. Past contributions focused on isolated and constrained factors whose core perspective was to create synthetic or re-recorded music corpus (e.g. GTSinger, M4Singer) and arbitrarily large-scale audio datasets (e.g. DISCO-10M and LAIONDISCO-12M) had been another focus for the community. Unfortunately, adoption of these datasets has been below substantial in the generative music community as these datasets fail to reflect real-world music and its flavour. Our dataset changes this narrative and provides a dataset that is constructed using actual popular music and world-renowned artists.
Authors: Alejandro Giraldo, Daniel Ruiz, Mariano Caruso, Javier Mancilla, Guido Bellomo
Abstract: Quantitative Structure-Activity Relationship (QSAR) modeling is a cornerstone of computational drug discovery. This research demonstrates the successful application of a Quantum Multiple Kernel Learning (QMKL) framework to enhance QSAR classification, showing a notable performance improvement over classical methods. We apply this methodology to a dataset for identifying DYRK1A kinase inhibitors. The workflow involves converting SMILES representations into numerical molecular descriptors, reducing dimensionality via Principal Component Analysis (PCA), and employing a Support Vector Machine (SVM) trained on an optimized combination of multiple quantum and classical kernels. By benchmarking the QMKL-SVM against a classical Gradient Boosting model, we show that the quantum-enhanced approach achieves a superior AUC score, highlighting its potential to provide a quantum advantage in challenging cheminformatics classification tasks.
Authors: Daqian Shao, Ashkan Soleymani, Francesco Quinzan, Marta Kwiatkowska
Abstract: Solving conditional moment restrictions (CMRs) is a key problem considered in statistics, causal inference, and econometrics, where the aim is to solve for a function of interest that satisfies some conditional moment equalities. Specifically, many techniques for causal inference, such as instrumental variable (IV) regression and proximal causal learning (PCL), are CMR problems. Most CMR estimators use a two-stage approach, where the first-stage estimation is directly plugged into the second stage to estimate the function of interest. However, naively plugging in the first-stage estimator can cause heavy bias in the second stage. This is particularly the case for recently proposed CMR estimators that use deep neural network (DNN) estimators for both stages, where regularisation and overfitting bias is present. We propose DML-CMR, a two-stage CMR estimator that provides an unbiased estimate with fast convergence rate guarantees. We derive a novel learning objective to reduce bias and develop the DML-CMR algorithm following the double/debiased machine learning (DML) framework. We show that our DML-CMR estimator can achieve the minimax optimal convergence rate of $O(N^{-1/2})$ under parameterisation and mild regularity conditions, where $N$ is the sample size. We apply DML-CMR to a range of problems using DNN estimators, including IV regression and proximal causal learning on real-world datasets, demonstrating state-of-the-art performance against existing CMR estimators and algorithms tailored to those problems.
Authors: Yunchi Lu, Youshan Miao, Cheng Tan, Peng Huang, Yi Zhu, Xian Zhang, Fan Yang
Abstract: Training large language models (LLMs) at scale requires parallel execution across thousands of devices, incurring enormous computational costs. Yet, these costly distributed trainings are rarely verified, leaving them prone to silent errors and potentially wasting millions of GPU hours. We introduce TrainVerify, a system for verifiable distributed training of LLMs. Given a deep learning model's logical specification as the ground truth, TrainVerify formally verifies that a distributed parallel execution plan is mathematically equivalent to it. Direct verification is notoriously difficult due to the sheer scale of LLMs which often involves billions of variables and highly intricate computation graphs. Therefore, TrainVerify introduces shape-reduction techniques and a stage-wise parallel verification algorithm that significantly reduces complexity while preserving formal correctness. TrainVerify scales to frontier LLMs, including the successful verification of the Llama3 (405B) and DeepSeek-V3 (671B) training plans.
Authors: Heng Zhang, Zikang Wan, Danilo Vasconcellos Vargas
Abstract: Human vision excels at segmenting visual cues without the need for explicit training, and it remains remarkably robust even as noise severity increases. In contrast, existing AI algorithms struggle to maintain accuracy under similar conditions. Here, we present SyncMapV2, the first to solve unsupervised segmentation with state-of-the-art robustness. SyncMapV2 exhibits a minimal drop in mIoU, only 0.01%, under digital corruption, compared to a 23.8% drop observed in SOTA methods. This superior performance extends across various types of corruption: noise (7.3% vs. 37.7%), weather (7.5% vs. 33.8%), and blur (7.0% vs. 29.5%). Notably, SyncMapV2 accomplishes this without any robust training, supervision, or loss functions. It is based on a learning paradigm that uses self-organizing dynamical equations combined with concepts from random networks. Moreover, unlike conventional methods that require re-initialization for each new input, SyncMapV2 adapts online, mimicking the continuous adaptability of human vision. Thus, we go beyond the accurate and robust results, and present the first algorithm that can do all the above online, adapting to input rather than re-initializing. In adaptability tests, SyncMapV2 demonstrates near-zero performance degradation, which motivates and fosters a new generation of robust and adaptive intelligence in the near future.