Authors: Jeesu Jung, Sangkeun Jung
Abstract: Curriculum learning for training LLMs requires a difficulty signal that aligns with reasoning while remaining scalable and interpretable. We propose a simple premise: tasks that demand deeper depth of thought for humans should also be harder for models. Accordingly, we define difficulty as depth of thought (DoT) and operationalize it by counting the discrete steps in a teacher model's reasoning trace (e.g., Chain-of-Thought). We then train with a shallow to deep curriculum ordered by this DoT and outline how to derive, validate, and schedule it at scale. Our position yields three testable hypotheses: (i) DoT correlates with conventional difficulty on reasoning benchmarks, (ii) DoT-ordered curricula outperform length- or judge-scored curricula under matched budgets, and (iii) the difficulty is robust across teacher models given light formatting controls. We propose an evaluation framework and discuss threats to validity (teacher style, length confounds) alongside practical mitigations. Taken together, we aim to move toward cognitively grounded, interpretable curricula for reasoning-centric training.
Authors: Rahmat K. Adesunkanmi, Alexander W. Brandt, Masoud Deylami, Gustavo A. Giraldo Echeverri, Hamidreza Karbasian, Adel Alaeddini
Abstract: Accurately predicting the drift (displacement) of leeway objects in maritime environments remains a critical challenge, particularly in time-sensitive scenarios such as search and rescue operations. In this study, we propose a multi-modal machine learning framework that integrates Sentence Transformer embeddings with attention-based sequence-to-sequence architectures to predict the drift of leeway objects in water. We begin by experimentally collecting environmental and physical data, including water current and wind velocities, object mass, and surface area, for five distinct leeway objects. Using simulated data from a Navier-Stokes-based model to train a convolutional neural network on geometrical image representations, we estimate drag and lift coefficients of the leeway objects. These coefficients are then used to derive the net forces responsible for driving the objects' motion. The resulting time series, comprising physical forces, environmental velocities, and object-specific features, combined with textual descriptions encoded via a language model, are inputs to attention-based sequence-to-sequence long-short-term memory and Transformer models, to predict future drift trajectories. We evaluate the framework across multiple time horizons ($1$, $3$, $5$, and $10$ seconds) and assess its generalization across different objects. We compare our approach against a fitted physics-based model and traditional machine learning methods, including recurrent neural networks and temporal convolutional neural networks. Our results show that these multi-modal models perform comparably to traditional models while also enabling longer-term forecasting in place of single-step prediction. Overall, our findings demonstrate the ability of a multi-modal modeling strategy to provide accurate and adaptable predictions of leeway object drift in dynamic maritime conditions.
Authors: Mateus A. Fernandes, Michael M. Furlanetti, Eduardo Gildin, Marcio A. Sampaio
Abstract: Forecasting production reliably and anticipating changes in the behavior of rock-fluid systems are the main challenges in petroleum reservoir engineering. This project proposes to deal with this problem through a data-driven approach and using machine learning methods. The objective is to develop a methodology to forecast production parameters based on simple data as produced and injected volumes and, eventually, gauges located in wells, without depending on information from geological models, fluid properties or details of well completions and flow systems. Initially, we performed relevance analyses of the production and injection variables, as well as conditioning the data to suit the problem. As reservoir conditions change over time, concept drift is a priority concern and require special attention to those observation windows and the periodicity of retraining, which are also objects of study. For the production forecasts, we study supervised learning methods, such as those based on regressions and Neural Networks, to define the most suitable for our application in terms of performance and complexity. In a first step, we evaluate the methodology using synthetic data generated from the UNISIM III compositional simulation model. Next, we applied it to cases of real plays in the Brazilian pre-salt. The expected result is the design of a reliable predictor for reproducing reservoir dynamics, with rapid response, capability of dealing with practical difficulties such as restrictions in wells and processing units, and that can be used in actions to support reservoir management, including the anticipation of deleterious behaviors, optimization of production and injection parameters and the analysis of the effects of probabilistic events, aiming to maximize oil recovery.
Authors: Md Sabbir Ahmed, Nova Ahmed
Abstract: Background: Existing robust, pervasive device-based systems developed in recent years to detect depression require data collected over a long period and may not be effective in cases where early detection is crucial. Objective: Our main objective was to develop a minimalistic system to identify depression using data retrieved in the fastest possible time. Methods: We developed a fast tool that retrieves the past 7 days' app usage data in 1 second (mean 0.31, SD 1.10 seconds). A total of 100 students from Bangladesh participated in our study, and our tool collected their app usage data. To identify depressed and nondepressed students, we developed a diverse set of ML models. We selected important features using the stable approach, along with 3 main types of feature selection (FS) approaches. Results: Leveraging only the app usage data retrieved in 1 second, our light gradient boosting machine model used the important features selected by the stable FS approach and correctly identified 82.4% (n=42) of depressed students (precision=75%, F1-score=78.5%). Moreover, after comprehensive exploration, we presented a parsimonious stacking model where around 5 features selected by the all-relevant FS approach Boruta were used in each iteration of validation and showed a maximum precision of 77.4% (balanced accuracy=77.9%). A SHAP analysis of our best models presented behavioral markers that were related to depression. Conclusions: Due to our system's fast and minimalistic nature, it may make a worthwhile contribution to identifying depression in underdeveloped and developing regions. In addition, our detailed discussion about the implication of our findings can facilitate the development of less resource-intensive systems to better understand students who are depressed.
Authors: Jueqi Wang, Zachary Jacokes, John Darrell Van Horn, Michael C. Schatz, Kevin A. Pelphrey, Archana Venkataraman
Abstract: While imaging-genetics holds great promise for unraveling the complex interplay between brain structure and genetic variation in neurological disorders, traditional methods are limited to simplistic linear models or to black-box techniques that lack interpretability. In this paper, we present NeuroPathX, an explainable deep learning framework that uses an early fusion strategy powered by cross-attention mechanisms to capture meaningful interactions between structural variations in the brain derived from MRI and established biological pathways derived from genetics data. To enhance interpretability and robustness, we introduce two loss functions over the attention matrix - a sparsity loss that focuses on the most salient interactions and a pathway similarity loss that enforces consistent representations across the cohort. We validate NeuroPathX on both autism spectrum disorder and Alzheimer's disease. Our results demonstrate that NeuroPathX outperforms competing baseline approaches and reveals biologically plausible associations linked to the disorder. These findings underscore the potential of NeuroPathX to advance our understanding of complex brain disorders. Code is available at https://github.com/jueqiw/NeuroPathX .
Authors: Wuxinlin Cheng, Yupeng Cao, Jinwen Wu, Koduvayur Subbalakshmi, Tian Han, Zhuo Feng
Abstract: Recent strides in pretrained transformer-based language models have propelled state-of-the-art performance in numerous NLP tasks. Yet, as these models grow in size and deployment, their robustness under input perturbations becomes an increasingly urgent question. Existing robustness methods often diverge between small-parameter and large-scale models (LLMs), and they typically rely on labor-intensive, sample-specific adversarial designs. In this paper, we propose a unified, local (sample-level) robustness framework (SALMAN) that evaluates model stability without modifying internal parameters or resorting to complex perturbation heuristics. Central to our approach is a novel Distance Mapping Distortion (DMD) measure, which ranks each sample's susceptibility by comparing input-to-output distance mappings in a near-linear complexity manner. By demonstrating significant gains in attack efficiency and robust training, we position our framework as a practical, model-agnostic tool for advancing the reliability of transformer-based NLP systems.
Authors: Mahishanka Withanachchi
Abstract: We introduce a unified framework for learning the spatio-temporal dynamics of vector valued functions by combining operator valued reproducing kernel Hilbert spaces (OV-RKHS) with kernel based Koopman operator methods. The approach enables nonparametric and data driven estimation of complex time evolving vector fields while preserving both spatial and temporal structure. We establish representer theorems for time dependent OV-RKHS interpolation, derive Sobolev type approximation bounds for smooth vector fields, and provide spectral convergence guarantees for kernel Koopman operator approximations. This framework supports efficient reduced order modeling and long term prediction of high dimensional nonlinear systems, offering theoretically grounded tools for forecasting, control, and uncertainty quantification in spatio-temporal machine learning.
Authors: Avinash Amballa
Abstract: Recent studies have demonstrated the effectiveness of position encoding in transformer architectures. By incorporating positional information, this approach provides essential guidance for modeling dependencies between elements across different sequence positions. We introduce CoPE (a lightweight Complex Positional Encoding), a novel architecture that leverages complex-valued encoding to encode both content and positional information. Our approach replaces traditional positional encodings with complex embeddings where the real part captures semantic content and the imaginary part encodes positional information. We introduce phase-aware attention in the first layer of the transformer model to capture position-dependent patterns, followed by standard attention layers for higher-levels. We show that CoPE doesn't exhibit long term decay and is compatible with linear attention. Experimental evaluation on the GLUE benchmark suggest that our approach achieves superior performance with less computational complexity, compared to RoPE, Sinusoidal and Learned positional encodings.
Authors: Yu Pan, Zhongze Cai, Guanting Chen, Huaiyang Zhong, Chonghuan Wang
Abstract: Direct Preference Optimization (DPO) has emerged as a simple and effective approach for aligning large language models (LLMs) with human preferences, bypassing the need for a learned reward model. Despite its growing adoption, a fundamental question remains open: what characteristics of preference data are most critical for DPO performance? In this work, we provide a systematic study of how preference data distribution influences DPO, from both theoretical and empirical perspectives. We show that the quality of chosen responses plays a dominant role in optimizing the DPO objective, while the quality of rejected responses may have relatively limited impact. Our theoretical analysis characterizes the optimal response distribution under DPO and reveals how contrastiveness between responses helps primarily by improving the chosen samples. We further study an online DPO setting and show it effectively reduces to supervised fine-tuning on the chosen responses. Extensive experiments across diverse tasks confirm our findings: improving the quality of chosen responses consistently boosts performance regardless of the quality of the rejected responses. We also investigate the benefit of mixing the on-policy data. Our results interpret the mechanism behind some widely adopted strategies and offer practical insights for constructing high-impact preference datasets for LLM alignment.
Authors: Zi Cai, Yu Liu, Zhiyao Luo, Tingting Zhu
Abstract: Digital healthcare systems have enabled the collection of mass healthcare data in electronic healthcare records (EHRs), allowing artificial intelligence solutions for various healthcare prediction tasks. However, existing studies often focus on isolated components of EHR data, limiting their predictive performance and interpretability. To address this gap, we propose ProtoEHR, an interpretable hierarchical prototype learning framework that fully exploits the rich, multi-level structure of EHR data to enhance healthcare predictions. More specifically, ProtoEHR models relationships within and across three hierarchical levels of EHRs: medical codes, hospital visits, and patients. We first leverage large language models to extract semantic relationships among medical codes and construct a medical knowledge graph as the knowledge source. Building on this, we design a hierarchical representation learning framework that captures contextualized representations across three levels, while incorporating prototype information within each level to capture intrinsic similarities and improve generalization. To perform a comprehensive assessment, we evaluate ProtoEHR in two public datasets on five clinically significant tasks, including prediction of mortality, prediction of readmission, prediction of length of stay, drug recommendation, and prediction of phenotype. The results demonstrate the ability of ProtoEHR to make accurate, robust, and interpretable predictions compared to baselines in the literature. Furthermore, ProtoEHR offers interpretable insights on code, visit, and patient levels to aid in healthcare prediction.
Authors: Rodrigo Tertulino
Abstract: High dropout and failure rates in distance education pose a significant challenge for academic institutions, making the proactive identification of at-risk students crucial for providing timely support. This study develops and evaluates a machine learning model based on early academic performance and digital engagement patterns from the large-scale OULAD dataset to predict student risk at a UK university. To address the practical challenges of data privacy and institutional silos that often hinder such initiatives, we implement the model using a Federated Learning (FL) framework. We compare model complexity (Logistic Regression vs. a Deep Neural Network) and data balancing. The final federated model demonstrates strong predictive capability, achieving an ROC AUC score of approximately 85% in identifying at-risk students. Our findings show that this federated approach provides a practical and scalable solution for institutions to build effective early-warning systems, enabling proactive student support while inherently respecting data privacy.
Authors: Yang Li, Hanjie Wang, Yuanzheng Li, Jiazheng Li, Zhaoyang Dong
Abstract: Wind power data often suffers from missing values due to sensor faults and unstable transmission at edge sites. While federated learning enables privacy-preserving collaboration without sharing raw data, it remains vulnerable to anomalous updates and privacy leakage during parameter exchange. These challenges are amplified in open industrial environments, necessitating zero-trust mechanisms where no participant is inherently trusted. To address these challenges, this work proposes ZTFed-MAS2S, a zero-trust federated learning framework that integrates a multi-head attention-based sequence-to-sequence imputation model. ZTFed integrates verifiable differential privacy with non-interactive zero-knowledge proofs and a confidentiality and integrity verification mechanism to ensure verifiable privacy preservation and secure model parameters transmission. A dynamic trust-aware aggregation mechanism is employed, where trust is propagated over similarity graphs to enhance robustness, and communication overhead is reduced via sparsity- and quantization-based compression. MAS2S captures long-term dependencies in wind power data for accurate imputation. Extensive experiments on real-world wind farm datasets validate the superiority of ZTFed-MAS2S in both federated learning performance and missing data imputation, demonstrating its effectiveness as a secure and efficient solution for practical applications in the energy sector.
Authors: Jarek Duda, Jagoda Bracha, Adrian Przybysz
Abstract: Evaluation of statistical dependencies between two data samples is a basic problem of data science/machine learning, and HSIC (Hilbert-Schmidt Information Criterion)~\cite{HSIC} is considered the state-of-art method. However, for size $n$ data sample it requires multiplication of $n\times n$ matrices, what currently needs $\sim O(n^{2.37})$ computational complexity~\cite{mult}, making it impractical for large data samples. We discuss HCR (Hierarchical Correlation Reconstruction) as its linear cost practical alternative of even higher dependence sensitivity in tests, and additionally providing actual joint distribution model by description of dependencies through features being mixed moments, starting with correlation and homoscedasticity, also allowing to approximate mutual information as just sum of squares of such nontrivial mixed moments between two data samples. Such single dependence describing feature is calculated in $O(n)$ linear time. Their number to test varies with dimension $d$ - requiring $O(d^2)$ for pairwise dependencies, $O(d^3)$ if wanting to also consider more subtle triplewise, and so on.
Authors: Weilin Cai, Le Qin, Shwai He, Junwei Cui, Ang Li, Jiayi Huang
Abstract: Mixture of Experts (MoE) has become a mainstream architecture for building Large Language Models (LLMs) by reducing per-token computation while enabling model scaling. It can be viewed as partitioning a large Feed-Forward Network (FFN) at the tensor level into fine-grained sub-FFNs, or experts, and activating only a sparse subset for each input. While this sparsity improves efficiency, MoE still faces substantial challenges due to their massive computational scale and unpredictable activation patterns. To enable efficient MoE deployment, we identify dual sparsity at the tensor and neuron levels in pre-trained MoE modules as a key factor for both accuracy and efficiency. Unlike prior work that increases tensor-level sparsity through finer-grained expert design during pre-training, we introduce post-training expert partitioning to induce such sparsity without retraining. This preserves the mathematical consistency of model transformations and enhances both efficiency and accuracy in subsequent fine-tuning and inference. Building upon this, we propose DualSparse-MoE, an inference system that integrates dynamic tensor-level computation dropping with static neuron-level reconstruction to deliver significant efficiency gains with minimal accuracy loss. Experimental results show that enforcing an approximate 25% drop rate with our approach reduces average accuracy by only 0.08%-0.28% across three prevailing MoE models, while nearly all degrees of computation dropping consistently yield proportional computational speedups. Furthermore, incorporating load-imbalance awareness into expert parallelism achieves a 1.41x MoE module speedup with just 0.5% average accuracy degradation.
Authors: Ricardo Borsoi, Konstantin Usevich, Marianne Clausel
Abstract: The groundbreaking performance of deep neural networks (NNs) promoted a surge of interest in providing a mathematical basis to deep learning theory. Low-rank tensor decompositions are specially befitting for this task due to their close connection to NNs and their rich theoretical results. Different tensor decompositions have strong uniqueness guarantees, which allow for a direct interpretation of their factors, and polynomial time algorithms have been proposed to compute them. Through the connections between tensors and NNs, such results supported many important advances in the theory of NNs. In this review, we show how low-rank tensor methods--which have been a core tool in the signal processing and machine learning communities--play a fundamental role in theoretically explaining different aspects of the performance of deep NNs, including their expressivity, algorithmic learnability and computational hardness, generalization, and identifiability. Our goal is to give an accessible overview of existing approaches (developed by different communities, ranging from computer science to mathematics) in a coherent and unified way, and to open a broader perspective on the use of low-rank tensor decompositions for the theory of deep NNs.
Authors: Andr\'e Quadros, Cassio Silva, Ronnie Alves
Abstract: This paper explores the combination of two intrinsic motivation strategies to improve the efficiency of reinforcement learning (RL) agents in environments with extreme sparse rewards, where traditional learning struggles due to infrequent positive feedback. We propose integrating Variational State as Intrinsic Reward (VSIMR), which uses Variational AutoEncoders (VAEs) to reward state novelty, with an intrinsic reward approach derived from Large Language Models (LLMs). The LLMs leverage their pre-trained knowledge to generate reward signals based on environment and goal descriptions, guiding the agent. We implemented this combined approach with an Actor-Critic (A2C) agent in the MiniGrid DoorKey environment, a benchmark for sparse rewards. Our empirical results show that this combined strategy significantly increases agent performance and sampling efficiency compared to using each strategy individually or a standard A2C agent, which failed to learn. Analysis of learning curves indicates that the combination effectively complements different aspects of the environment and task: VSIMR drives exploration of new states, while the LLM-derived rewards facilitate progressive exploitation towards goals.
Authors: Quanlin Chen, Yiyu Chen, Jing Huo, Tianyu Ding, Yang Gao, Yuetong Chen
Abstract: Bayesian Optimization (BO) has been widely applied to optimize expensive black-box functions while retaining sample efficiency. However, scaling BO to high-dimensional spaces remains challenging. Existing literature proposes performing standard BO in multiple local trust regions (TuRBO) for heterogeneous modeling of the objective function and avoiding over-exploration. Despite its advantages, using local Gaussian Processes (GPs) reduces sampling efficiency compared to a global GP. To enhance sampling efficiency while preserving heterogeneous modeling, we propose to construct multiple local quadratic models using gradients and Hessians from a global GP, and select new sample points by solving the bound-constrained quadratic program. Additionally, we address the issue of vanishing gradients of GPs in high-dimensional spaces. We provide a convergence analysis and demonstrate through experimental results that our method enhances the efficacy of TuRBO and outperforms a wide range of high-dimensional BO techniques on synthetic functions and real-world applications.
Authors: Fu Teng, Miao Pan, Xuhong Zhang, Zhezhi He, Yiyao Yang, Xinyi Chai, Mengnan Qi, Liqiang Lu, Jianwei Yin
Abstract: Recent advancements in code generation have shown remarkable success across software domains, yet hardware description languages (HDLs) such as Verilog remain underexplored due to their concurrency semantics, syntactic rigidity, and simulation complexity. In this work, we address these challenges by introducing a reinforcement learning (RL) framework tailored for Verilog code generation. We first construct Veribench-53K, a high-quality dataset curated from over 700K Verilog problems, enriched with structured prompts, complexity labels, and diverse testbenches. To tackle the problem of sparse and noisy reward signals, we propose a Trace-back based Rescore mechanism that leverages reasoning paths and iterative refinement to enhance feedback reliability and support reward model training. Furthermore, to mitigate catastrophic forgetting and overfitting during RL fine-tuning, we introduce a sample-balanced weighting strategy that adaptively balances learning dynamics based on reward-probability distributions. These innovations are integrated into an iterative RL pipeline that co-evolves the policy and reward models. In contrast to recent work such as CraftRTL, which relies on large-scale closed-source model distillation, and DeepSeek-style approaches that struggle with sparse feedback, our method demonstrates superior performance using a smaller but high-quality dataset combined with RL optimization. Experiments on Verilog generation tasks demonstrate state-of-the-art performance, with substantial gains in test pass rate, functional correctness, and compilation robustness. Our findings highlight the potential of RL-driven approaches for structured code generation in hardware-centric domains. VERIRL is publicly available at https://github.com/omniAI-Lab/VeriRL.
Authors: Bahareh Golchin, Banafsheh Rekabdar, Kunpeng Liu
Abstract: Anomaly detection in time series data is important for applications in finance, healthcare, sensor networks, and industrial monitoring. Traditional methods usually struggle with limited labeled data, high false-positive rates, and difficulty generalizing to novel anomaly types. To overcome these challenges, we propose a reinforcement learning-based framework that integrates dynamic reward shaping, Variational Autoencoder (VAE), and active learning, called DRTA. Our method uses an adaptive reward mechanism that balances exploration and exploitation by dynamically scaling the effect of VAE-based reconstruction error and classification rewards. This approach enables the agent to detect anomalies effectively in low-label systems while maintaining high precision and recall. Our experimental results on the Yahoo A1 and Yahoo A2 benchmark datasets demonstrate that the proposed method consistently outperforms state-of-the-art unsupervised and semi-supervised approaches. These findings show that our framework is a scalable and efficient solution for real-world anomaly detection tasks.
Authors: Andreza M. C. Falcao, Filipe R. Cordeiro
Abstract: Machine Unlearning (MU) aims to remove the influence of specific data from a trained model while preserving its performance on the remaining data. Although a few works suggest connections between memorisation and augmentation, the role of systematic augmentation design in MU remains under-investigated. In this work, we investigate the impact of different data augmentation strategies on the performance of unlearning methods, including SalUn, Random Label, and Fine-Tuning. Experiments conducted on CIFAR-10 and CIFAR-100, under varying forget rates, show that proper augmentation design can significantly improve unlearning effectiveness, reducing the performance gap to retrained models. Results showed a reduction of up to 40.12% of the Average Gap unlearning Metric, when using TrivialAug augmentation. Our results suggest that augmentation not only helps reduce memorization but also plays a crucial role in achieving privacy-preserving and efficient unlearning.
Authors: Yifeng Peng, Xinyi Li, Zhemin Zhang, Samuel Yen-Chi Chen, Zhiding Liang, Ying Wang
Abstract: Variational Quantum Algorithms (VQAs) have gained prominence as a viable framework for exploiting near-term quantum devices in applications ranging from optimization and chemistry simulation to machine learning. However, the effectiveness of VQAs is often constrained by the so-called barren plateau problem, wherein gradients diminish exponentially as system size or circuit depth increases, thereby hindering training. In this work, we propose a reinforcement learning (RL)-based initialization strategy to alleviate the barren plateau issue by reshaping the initial parameter landscape to avoid regions prone to vanishing gradients. In particular, we explore several RL algorithms (Deterministic Policy Gradient, Soft Actor-Critic, and Proximal Policy Optimization, etc.) to generate the circuit parameters (treated as actions) that minimize the VQAs cost function before standard gradient-based optimization. By pre-training with RL in this manner, subsequent optimization using methods such as gradient descent or Adam proceeds from a more favorable initial state. Extensive numerical experiments under various noise conditions and tasks consistently demonstrate that the RL-based initialization method significantly enhances both convergence speed and final solution quality. Moreover, comparisons among different RL algorithms highlight that multiple approaches can achieve comparable performance gains, underscoring the flexibility and robustness of our method. These findings shed light on a promising avenue for integrating machine learning techniques into quantum algorithm design, offering insights into how RL-driven parameter initialization can accelerate the scalability and practical deployment of VQAs. Opening up a promising path for the research community in machine learning for quantum, especially barren plateau problems in VQAs.
Authors: Anastasis Kratsios, Dennis Zvigelsky, Bradd Hart
Abstract: A main open question in contemporary AI research is quantifying the forms of reasoning neural networks can perform when perfectly trained. This paper answers this by interpreting reasoning tasks as circuit emulation, where the gates define the type of reasoning; e.g. Boolean gates for predicate logic, tropical circuits for dynamic programming, arithmetic and analytic gates for symbolic mathematical representation, and hybrids thereof for deeper reasoning; e.g. higher-order logic. We present a systematic meta-algorithm that converts essentially any circuit into a feedforward neural network (NN) with ReLU activations by iteratively replacing each gate with a canonical ReLU MLP emulator. We show that, on any digital computer, our construction emulates the circuit exactly--no approximation, no rounding, modular overflow included--demonstrating that no reasoning task lies beyond the reach of neural networks. The number of neurons in the resulting network (parametric complexity) scales with the circuit's complexity, and the network's computational graph (structure) mirrors that of the emulated circuit. This formalizes the folklore that NNs networks trade algorithmic run-time (circuit runtime) for space complexity (number of neurons). We derive a range of applications of our main result, from emulating shortest-path algorithms on graphs with cubic--size NNs, to simulating stopped Turing machines with roughly quadratically--large NNs, and even the emulation of randomized Boolean circuits. Lastly, we demonstrate that our result is strictly more powerful than a classical universal approximation theorem: any universal function approximator can be encoded as a circuit and directly emulated by a NN.
Authors: Jun Hou, Le Wang, Xuan Wang
Abstract: Mixture-of-Experts (MoE) models have become increasingly powerful in multimodal learning by enabling modular specialization across modalities. However, their effectiveness remains unclear when additional modalities introduce more noise than complementary information. Existing approaches, such as the Partial Information Decomposition, struggle to scale beyond two modalities and lack the resolution needed for instance-level control. We propose Beyond Two-modality Weighting (BTW), a bi-level, non-parametric weighting framework that combines instance-level Kullback-Leibler (KL) divergence and modality-level mutual information (MI) to dynamically adjust modality importance during training. Our method does not require additional parameters and can be applied to an arbitrary number of modalities. Specifically, BTW computes per-example KL weights by measuring the divergence between each unimodal and the current multimodal prediction, and modality-wide MI weights by estimating global alignment between unimodal and multimodal outputs. Extensive experiments on sentiment regression and clinical classification demonstrate that our method significantly improves regression performance and multiclass classification accuracy.
Authors: {\L}ukasz Janisi\'ow, Marek Kocha\'nczyk, Bartosz Zieli\'nski, Tomasz Danel
Abstract: Molecular property prediction is a crucial task that guides the design of new compounds, including drugs and materials. While explainable artificial intelligence methods aim to scrutinize model predictions by identifying influential molecular substructures, many existing approaches rely on masking strategies that remove either atoms or atom-level features to assess importance via fidelity metrics. These methods, however, often fail to adhere to the underlying molecular distribution and thus yield unintuitive explanations. In this work, we propose counterfactual masking, a novel framework that replaces masked substructures with chemically reasonable fragments sampled from generative models trained to complete molecular graphs. Rather than evaluating masked predictions against implausible zeroed-out baselines, we assess them relative to counterfactual molecules drawn from the data distribution. Our method offers two key benefits: (1) molecular realism underpinning robust and distribution-consistent explanations, and (2) meaningful counterfactuals that directly indicate how structural modifications may affect predicted properties. We demonstrate that counterfactual masking is well-suited for benchmarking model explainers and yields more actionable insights across multiple datasets and property prediction tasks. Our approach bridges the gap between explainability and molecular design, offering a principled and generative path toward explainable machine learning in chemistry.
Authors: Levi Rauchwerger, Ron Levie
Abstract: A recent paper, ``A Graphon-Signal Analysis of Graph Neural Networks'', by Levie, analyzed message passing graph neural networks (MPNNs) by embedding the input space of MPNNs, i.e., attributed graphs (graph-signals), to a space of attributed graphons (graphon-signals). Based on extensions of standard results in graphon analysis to graphon-signals, the paper proved a generalization bound and a sampling lemma for MPNNs. However, there are some missing ingredients in that paper, limiting its applicability in practical settings of graph machine learning. In the current paper, we introduce several refinements and extensions to existing results that address these shortcomings. In detail, 1) we extend the main results in the paper to graphon-signals with multidimensional signals (rather than 1D signals), 2) we extend the Lipschitz continuity to MPNNs with readout with respect to cut distance (rather than MPNNs without readout with respect to cut metric), 3) we improve the generalization bound by utilizing robustness-type generalization bounds, and 4) we extend the analysis to non-symmetric graphons and kernels.
Authors: Hao Zhou, Sibo Cheng
Abstract: Data-driven methods are emerging as efficient alternatives to traditional numerical forecasting, offering fast inference and lower computational cost. Yet, for complex systems, long-term accuracy often deteriorates due to error accumulation, and autoregressive training (though effective) demands large GPU memory and may sacrifice short-term performance. We propose the Stochastic PushForward (SPF) framework, which retains one-step-ahead training while enabling multi-step learning. SPF builds a supplementary dataset from model predictions and combines it with ground truth via a stochastic acquisition strategy, balancing short- and long-term performance while reducing overfitting. Multi-step predictions are precomputed between epochs, keeping memory usage stable without storing full unrolled sequences. Experiments on the Burgers' equation and the Shallow Water benchmark show that SPF achieves higher long-term accuracy than autoregressive methods while lowering memory requirements, making it promising for resource-limited and complex simulations.
Authors: Darin Tsui, Kunal Talreja, Amirali Aghazadeh
Abstract: Predicting protein function from amino acid sequence remains a central challenge in data-scarce (low-$N$) regimes, limiting machine learning-guided protein design when only small amounts of assay-labeled sequence-function data are available. Protein language models (pLMs) have advanced the field by providing evolutionary-informed embeddings and sparse autoencoders (SAEs) have enabled decomposition of these embeddings into interpretable latent variables that capture structural and functional features. However, the effectiveness of SAEs for low-$N$ function prediction and protein design has not been systematically studied. Herein, we evaluate SAEs trained on fine-tuned ESM2 embeddings across diverse fitness extrapolation and protein engineering tasks. We show that SAEs, with as few as 24 sequences, consistently outperform or compete with their ESM2 baselines in fitness prediction, indicating that their sparse latent space encodes compact and biologically meaningful representations that generalize more effectively from limited data. Moreover, steering predictive latents exploits biological motifs in pLM representations, yielding top-fitness variants in 83% of cases compared to designing with ESM2 alone.
Authors: Mohammadreza Ghaffarzadeh-Esfahani, Ali Motahharynia, Nahid Yousefian, Navid Mazrouei, Jafar Ghaisari, Yousof Gheisari
Abstract: Drug discovery is a complex and resource-intensive process, making early prediction of approval outcomes critical for optimizing research investments. While classical machine learning and deep learning methods have shown promise in drug approval prediction, their limited interpretability constraints their impact. Here, we present DrugReasoner, a reasoning-based large language model (LLM) built on the LLaMA architecture and fine-tuned with group relative policy optimization (GRPO) to predict the likelihood of small-molecule approval. DrugReasoner integrates molecular descriptors with comparative reasoning against structurally similar approved and unapproved compounds, generating predictions alongside step-by-step rationales and confidence scores. DrugReasoner achieved robust performance with an AUC of 0.732 and an F1 score of 0.729 on the validation set and 0.725 and 0.718 on the test set, respectively. These results outperformed conventional baselines, including logistic regression, support vector machine, and k-nearest neighbors and had competitive performance relative to XGBoost. On an external independent dataset, DrugReasoner outperformed both baseline and the recently developed ChemAP model, achieving an AUC of 0.728 and an F1-score of 0.774, while maintaining high precision and balanced sensitivity, demonstrating robustness in real-world scenarios. These findings demonstrate that DrugReasoner not only delivers competitive predictive accuracy but also enhances transparency through its reasoning outputs, thereby addressing a key bottleneck in AI-assisted drug discovery. This study highlights the potential of reasoning-augmented LLMs as interpretable and effective tools for pharmaceutical decision-making.
Authors: Jingkai He, Tianjian Li, Erhu Feng, Dong Du, Qian Liu, Tao Liu, Yubin Xia, Haibo Chen
Abstract: With the rapid advancement of large language models (LLMs), reinforcement learning (RL) has emerged as a pivotal methodology for enhancing the reasoning capabilities of LLMs. Unlike traditional pre-training approaches, RL encompasses multiple stages: rollout, reward, and training, which necessitates collaboration among various worker types. However, current RL systems continue to grapple with substantial GPU underutilization, due to two primary factors: (1) The rollout stage dominates the overall RL process due to test-time scaling; (2) Imbalances in rollout lengths (within the same batch) result in GPU bubbles. While prior solutions like asynchronous execution and truncation offer partial relief, they may compromise training accuracy for efficiency. Our key insight stems from a previously overlooked observation: rollout responses exhibit remarkable similarity across adjacent training epochs. Based on the insight, we introduce RhymeRL, an LLM RL system designed to accelerate RL training with two key innovations. First, to enhance rollout generation, we present HistoSpec, a speculative decoding inference engine that utilizes the similarity of historical rollout token sequences to obtain accurate drafts. Second, to tackle rollout bubbles, we introduce HistoPipe, a two-tier scheduling strategy that leverages the similarity of historical rollout distributions to balance workload among rollout workers. We have evaluated RhymeRL within a real production environment, demonstrating scalability from dozens to thousands of GPUs. Experimental results demonstrate that RhymeRL achieves a 2.6x performance improvement over existing methods, without compromising accuracy or modifying the RL paradigm.
Authors: Zhao-Rong Lai, Haisheng Yang
Abstract: The principal portfolio approach is an emerging method in signal-based trading. However, these principal portfolios may not be diversified to explore the key features of the prediction matrix or robust to different situations. To address this problem, we propose a novel linear trading position with sparse spectrum that can explore a larger spectral region of the prediction matrix. We also develop a Krasnosel'ski\u \i-Mann fixed-point algorithm to optimize this trading position, which possesses the descent property and achieves a linear convergence rate in the objective value. This is a new theoretical result for this type of algorithms. Extensive experiments show that the proposed method achieves good and robust performance in various situations.
Authors: Weide Liu, Xiaoyang Zhong, Lu Wang, Jingwen Hou, Yuemei Luo, Jiebin Yan, Yuming Fang
Abstract: Unsupervised domain adaptation methods seek to generalize effectively on unlabeled test data, especially when encountering the common challenge in time series data that distribution shifts occur between training and testing datasets. In this paper, we propose incorporating multi-scale feature extraction and uncertainty estimation to improve the model's generalization and robustness across domains. Our approach begins with a multi-scale mixed input architecture that captures features at different scales, increasing training diversity and reducing feature discrepancies between the training and testing domains. Based on the mixed input architecture, we further introduce an uncertainty awareness mechanism based on evidential learning by imposing a Dirichlet prior on the labels to facilitate both target prediction and uncertainty estimation. The uncertainty awareness mechanism enhances domain adaptation by aligning features with the same labels across different domains, which leads to significant performance improvements in the target domain. Additionally, our uncertainty-aware model demonstrates a much lower Expected Calibration Error (ECE), indicating better-calibrated prediction confidence. Our experimental results show that this combined approach of mixed input architecture with the uncertainty awareness mechanism achieves state-of-the-art performance across multiple benchmark datasets, underscoring its effectiveness in unsupervised domain adaptation for time series data.
Authors: Yue Jiang, Chenxi Liu, Yile Chen, Qin Chao, Shuai Liu, Gao Cong
Abstract: Urban forecasting models often face a severe data imbalance problem: only a few cities have dense, long-span records, while many others expose short or incomplete histories. Direct transfer from data-rich to data-scarce cities is unreliable because only a limited subset of source patterns truly benefits the target domain, whereas indiscriminate transfer risks introducing noise and negative transfer. We present STRATA-TS (Selective TRAnsfer via TArget-aware retrieval for Time Series), a framework that combines domain-adapted retrieval with reasoning-capable large models to improve forecasting in scarce data regimes. STRATA-TS employs a patch-based temporal encoder to identify source subsequences that are semantically and dynamically aligned with the target query. These retrieved exemplars are then injected into a retrieval-guided reasoning stage, where an LLM performs structured inference over target inputs and retrieved support. To enable efficient deployment, we distill the reasoning process into a compact open model via supervised fine-tuning. Extensive experiments on three parking availability datasets across Singapore, Nottingham, and Glasgow demonstrate that STRATA-TS consistently outperforms strong forecasting and transfer baselines, while providing interpretable knowledge transfer pathways.
Authors: Ifrah Tariq, Ernest Fraenkel
Abstract: Immune checkpoint inhibitors (ICIs) have transformed cancer treatment, yet patient responses remain highly variable, and the biological mechanisms underlying resistance are poorly understood. While machine learning models hold promise for predicting responses to ICIs, most existing methods lack interpretability and do not effectively leverage the biological structure inherent to multi-omics data. Here, we introduce the Biologically Disentangled Variational Autoencoder (BDVAE), a deep generative model that integrates transcriptomic and genomic data through modality- and pathway-specific encoders. Unlike existing rigid, pathway-informed models, BDVAE employs a modular encoder architecture combined with variational inference to learn biologically meaningful latent features associated with immune, genomic, and metabolic processes. Applied to a pan-cancer cohort of 366 patients across four cancer types treated with ICIs, BDVAE accurately predicts treatment response (AUC-ROC = 0.94 on unseen test data) and uncovers critical resistance mechanisms, including immune suppression, metabolic shifts, and neuronal signaling. Importantly, BDVAE reveals that resistance spans a continuous biological spectrum rather than strictly binary states, reflecting gradations of tumor dysfunction. Several latent features correlate with survival outcomes and known clinical subtypes, demonstrating BDVAE's capability to generate interpretable, clinically relevant insights. These findings underscore the value of biologically structured machine learning in elucidating complex resistance patterns and guiding precision immunotherapy strategies.
Authors: Xiaoliang Chen, Xin Yu, Le Chang, Teng Jing, Jiashuai He, Ze Wang, Yangjun Luo, Xingyu Chen, Jiayue Liang, Yuchen Wang, Jiaying Xie
Abstract: Information asymmetry in financial markets, often amplified by strategically crafted corporate narratives, undermines the effectiveness of conventional textual analysis. We propose a novel multimodal framework for financial risk assessment that integrates textual sentiment with paralinguistic cues derived from executive vocal tract dynamics in earnings calls. Central to this framework is the Physics-Informed Acoustic Model (PIAM), which applies nonlinear acoustics to robustly extract emotional signatures from raw teleconference sound subject to distortions such as signal clipping. Both acoustic and textual emotional states are projected onto an interpretable three-dimensional Affective State Label (ASL) space-Tension, Stability, and Arousal. Using a dataset of 1,795 earnings calls (approximately 1,800 hours), we construct features capturing dynamic shifts in executive affect between scripted presentation and spontaneous Q&A exchanges. Our key finding reveals a pronounced divergence in predictive capacity: while multimodal features do not forecast directional stock returns, they explain up to 43.8% of the out-of-sample variance in 30-day realized volatility. Importantly, volatility predictions are strongly driven by emotional dynamics during executive transitions from scripted to spontaneous speech, particularly reduced textual stability and heightened acoustic instability from CFOs, and significant arousal variability from CEOs. An ablation study confirms that our multimodal approach substantially outperforms a financials-only baseline, underscoring the complementary contributions of acoustic and textual modalities. By decoding latent markers of uncertainty from verifiable biometric signals, our methodology provides investors and regulators a powerful tool for enhancing market interpretability and identifying hidden corporate uncertainty.
Authors: Gang Hu, Yinglei Teng, Pengfei Wu, Nan Wang
Abstract: As FMs drive progress toward Artificial General Intelligence (AGI), fine-tuning them under privacy and resource constraints has become increasingly critical particularly when highquality training data resides on distributed edge devices. Federated Learning (FL) offers a compelling solution through Federated Fine-Tuning (FFT), which enables collaborative model adaptation without sharing raw data. Recent approaches incorporate Parameter-Efficient Fine-Tuning (PEFT) techniques such as Low Rank Adaptation (LoRA) to reduce computational overhead. However, LoRA-based FFT faces two major limitations in heterogeneous FL environments: structural incompatibility across clients with varying LoRA configurations and limited adaptability to non-IID data distributions, which hinders convergence and generalization. To address these challenges, we propose FFT MoE, a novel FFT framework that replaces LoRA with sparse Mixture of Experts (MoE) adapters. Each client trains a lightweight gating network to selectively activate a personalized subset of experts, enabling fine-grained adaptation to local resource budgets while preserving aggregation compatibility. To further combat the expert load imbalance caused by device and data heterogeneity, we introduce a heterogeneity-aware auxiliary loss that dynamically regularizes the routing distribution to ensure expert diversity and balanced utilization. Extensive experiments spanning both IID and non-IID conditions demonstrate that FFT MoE consistently outperforms state of the art FFT baselines in generalization performance and training efficiency.
Authors: Yuechun Gu, Jiajie He, Keke Chen
Abstract: Approximate machine unlearning aims to remove the effect of specific data from trained models to ensure individuals' privacy. Existing methods focus on the removed records and assume the retained ones are unaffected. However, recent studies on the \emph{privacy onion effect} indicate this assumption might be incorrect. Especially when the model is differentially private, no study has explored whether the retained ones still meet the differential privacy (DP) criterion under existing machine unlearning methods. This paper takes a holistic approach to auditing both unlearned and retained samples' privacy risks after applying approximate unlearning algorithms. We propose the privacy criteria for unlearned and retained samples, respectively, based on the perspectives of DP and membership inference attacks (MIAs). To make the auditing process more practical, we also develop an efficient MIA, A-LiRA, utilizing data augmentation to reduce the cost of shadow model training. Our experimental findings indicate that existing approximate machine unlearning algorithms may inadvertently compromise the privacy of retained samples for differentially private models, and we need differentially private unlearning algorithms. For reproducibility, we have pubished our code: https://anonymous.4open.science/r/Auditing-machine-unlearning-CB10/README.md
URLs: https://anonymous.4open.science/r/Auditing-machine-unlearning-CB10/README.md
Authors: Taishi Nakamura, Satoki Ishikawa, Masaki Kawamura, Takumi Okamoto, Daisuke Nohara, Jun Suzuki, Rio Yokota
Abstract: Empirical scaling laws have driven the evolution of large language models (LLMs), yet their coefficients shift whenever the model architecture or data pipeline changes. Mixture-of-Experts (MoE) models, now standard in state-of-the-art systems, introduce a new sparsity dimension that current dense-model frontiers overlook. We investigate how MoE sparsity influences two distinct capability regimes: memorization and reasoning. We train families of MoE Transformers that systematically vary total parameters, active parameters, and top-$k$ routing while holding the compute budget fixed. For every model we record pre-training loss, downstream task loss, and task accuracy, allowing us to separate the train-test generalization gap from the loss-accuracy gap. Memorization benchmarks improve monotonically with total parameters, mirroring training loss. By contrast, reasoning performance saturates and can even regress despite continued gains in both total parameters and training loss. Altering top-$k$ alone has little effect when active parameters are constant, and classic hyperparameters such as learning rate and initialization modulate the generalization gap in the same direction as sparsity. Neither post-training reinforcement learning (GRPO) nor extra test-time compute rescues the reasoning deficit of overly sparse models. Our model checkpoints, code and logs are open-source at https://github.com/rioyokotalab/optimal-sparsity.
Authors: Chufan Gao, Jintai Chen, Jimeng Sun
Abstract: Automated tabular understanding and reasoning are essential tasks for data scientists. Recently, Large language models (LLMs) have become increasingly prevalent in tabular reasoning tasks. Previous work focuses on (1) finetuning LLMs using labeled data or (2) Training-free prompting LLM agents using chain-of-thought (CoT). Finetuning offers dataset-specific learning at the cost of generalizability. Training-free prompting is highly generalizable but does not take full advantage of training data. In this paper, we propose a novel prompting-based reasoning approach, Learn then Retrieve: LRTab, which integrates the benefits of both by retrieving relevant information learned from training data. We first use prompting to obtain CoT responses over the training data. For incorrect CoTs, we prompt the LLM to predict Prompt Conditions to avoid the error, learning insights from the data. We validate the effectiveness of Prompt Conditions using validation data. Finally, at inference time, we retrieve the most relevant Prompt Conditions for additional context for table understanding. We provide comprehensive experiments on WikiTQ and Tabfact, showing that LRTab is interpretable, cost-efficient, and can outperform previous baselines in tabular reasoning.
Authors: Hejiang Cai, Di Wu, Ji Xu, Xiang Liu, Yiziting Zhu, Xin Shu, Yujie Li, Bin Yi
Abstract: Sepsis is a life threatening condition that requires timely detection in intensive care settings. Traditional machine learning approaches, including Naive Bayes, Support Vector Machine (SVM), Random Forest, and XGBoost, often rely on manual feature engineering and struggle with irregular, incomplete time-series data commonly present in electronic health records. We introduce an end-to-end deep learning framework integrating an unsupervised autoencoder for automatic feature extraction with a multilayer perceptron classifier for binary sepsis risk prediction. To enhance clinical applicability, we implement a customized down sampling strategy that extracts high information density segments during training and a non-overlapping dynamic sliding window mechanism for real-time inference. Preprocessed time series data are represented as fixed dimension vectors with explicit missingness indicators, mitigating bias and noise. We validate our approach on three ICU cohorts. Our end-to-end model achieves accuracies of 74.6 percent, 80.6 percent, and 93.5 percent, respectively, consistently outperforming traditional machine learning baselines. These results demonstrate the framework's superior robustness, generalizability, and clinical utility for early sepsis detection across heterogeneous ICU environments.
Authors: V. S. Usatyuk, D. A. Sapoznikov, S. I. Egorov
Abstract: We present a unified framework combining statistical physics, coding theory, and algebraic topology for efficient multi-class image classification. High-dimensional feature vectors from a frozen MobileNetV2 backbone are interpreted as spins on a sparse Multi-Edge Type quasi-cyclic LDPC (MET-QC-LDPC) graph, forming a Random-Bond Ising Model (RBIM). We operate this RBIM at its Nishimori temperature, $\beta_N$, where the smallest eigenvalue of the Bethe-Hessian matrix vanishes, maximizing class separability. Our theoretical contribution establishes a correspondence between local trapping sets in the code's graph and topological invariants (Betti numbers, bordism classes) of the feature manifold. A practical algorithm estimates $\beta_N$ efficiently with a quadratic interpolant and Newton correction, achieving a six-fold speed-up over bisection. Guided by topology, we design spherical and toroidal MET-QC-LDPC graph ensembles, using permanent bounds to suppress harmful trapping sets. This compresses 1280-dimensional features to 32 or 64 dimensions for ImageNet-10 and -100 subsets. Despite massive compression (40x fewer parameters), we achieve 98.7% accuracy on ImageNet-10 and 82.7% on ImageNet-100, demonstrating that topology-guided graph design yields highly efficient, physics-inspired embeddings with state-of-the-art performance.
Authors: Yi Liu, Hongji Zhang, Yiwen Wang, Dimitris Tsaras, Lei Chen, Mingxuan Yuan, Qiang Xu
Abstract: Estimating the quality of register transfer level (RTL) designs is crucial in the electronic design automation (EDA) workflow, as it enables instant feedback on key metrics like area and delay without the need for time-consuming logic synthesis. While recent approaches have leveraged large language models (LLMs) to derive embeddings from RTL code and achieved promising results, they overlook the structural semantics essential for accurate quality estimation. In contrast, the control data flow graph (CDFG) view exposes the design's structural characteristics more explicitly, offering richer cues for representation learning. In this work, we introduce a novel structure-aware graph self-supervised learning framework, StructRTL, for improved RTL design quality estimation. By learning structure-informed representations from CDFGs, our method significantly outperforms prior art on various quality estimation tasks. To further boost performance, we incorporate a knowledge distillation strategy that transfers low-level insights from post-mapping netlists into the CDFG predictor. Experiments show that our approach establishes new state-of-the-art results, demonstrating the effectiveness of combining structural learning with cross-stage supervision.
Authors: Enrique M\'armol Campos, Aurora Gonz\'alez Vidal, Jos\'e Luis Hern\'andez Ramos, Antonio Skarmeta
Abstract: Federated Learning (FL) has become a powerful technique for training Machine Learning (ML) models in a decentralized manner, preserving the privacy of the training datasets involved. However, the decentralized nature of FL limits the visibility of the training process, relying heavily on the honesty of participating clients. This assumption opens the door to malicious third parties, known as Byzantine clients, which can poison the training process by submitting false model updates. Such malicious clients may engage in poisoning attacks, manipulating either the dataset or the model parameters to induce misclassification. In response, this study introduces FLAegis, a two-stage defensive framework designed to identify Byzantine clients and improve the robustness of FL systems. Our approach leverages symbolic time series transformation (SAX) to amplify the differences between benign and malicious models, and spectral clustering, which enables accurate detection of adversarial behavior. Furthermore, we incorporate a robust FFT-based aggregation function as a final layer to mitigate the impact of those Byzantine clients that manage to evade prior defenses. We rigorously evaluate our method against five poisoning attacks, ranging from simple label flipping to adaptive optimization-based strategies. Notably, our approach outperforms state-of-the-art defenses in both detection precision and final model accuracy, maintaining consistently high performance even under strong adversarial conditions.
Authors: Enoch H. Kang, Kyoungseok Jang
Abstract: Offline reinforcement learning and offline inverse reinforcement learning aim to recover near-optimal value functions or reward models from a fixed batch of logged trajectories, yet current practice still struggles to enforce Bellman consistency. Bellman residual minimization (BRM) has emerged as an attractive remedy, as a globally convergent stochastic gradient descent-ascent based method for BRM has been recently discovered. However, its statistical behavior in the offline setting remains largely unexplored. In this paper, we close this statistical gap. Our analysis introduces a single Lyapunov potential that couples SGDA runs on neighbouring datasets and yields an O(1/n) on-average argument-stability bound-doubling the best known sample-complexity exponent for convex-concave saddle problems. The same stability constant translates into the O(1/n) excess risk bound for BRM, without variance reduction, extra regularization, or restrictive independence assumptions on minibatch sampling. The results hold for standard neural-network parameterizations and minibatch SGD.
Authors: Jiajun Li, Ran Hou, Yu Ding, Yixuan Li, Shisi Guan, Jiahui Duan, Xiongwei Han, Tao Zhong, Vincent Chau, Weiwei Wu, Wanyuan Wang
Abstract: Model reduction, which aims to learn a simpler model of the original mixed integer linear programming (MILP), can solve large-scale MILP problems much faster. Most existing model reduction methods are based on variable reduction, which predicts a solution value for a subset of variables. From a dual perspective, constraint reduction that transforms a subset of inequality constraints into equalities can also reduce the complexity of MILP, but has been largely ignored. Therefore, this paper proposes a novel constraint-based model reduction approach for the MILP. Constraint-based MILP reduction has two challenges: 1) which inequality constraints are critical such that reducing them can accelerate MILP solving while preserving feasibility, and 2) how to predict these critical constraints efficiently. To identify critical constraints, we first label these tight-constraints at the optimal solution as potential critical constraints and design a heuristic rule to select a subset of critical tight-constraints. To learn the critical tight-constraints, we propose a multi-modal representation technique that leverages information from both instance-level and abstract-level MILP formulations. The experimental results show that, compared to the state-of-the-art methods, our method improves the quality of the solution by over 50\% and reduces the computation time by 17.47\%.
Authors: Zihao Huang, Yu Bao, Qiyang Min, Siyan Chen, Ran Guo, Hongzhi Huang, Defa Zhu, Yutao Zeng, Banggu Wu, Xun Zhou, Siyuan Qiao
Abstract: While Mixture of Experts (MoE) models achieve remarkable efficiency by activating only subsets of parameters, they suffer from high memory access costs during inference. Memory-layer architectures offer an appealing alternative with very few memory access, but previous attempts like UltraMem have only matched the performance of 2-expert MoE models, falling significantly short of state-of-the-art 8-expert configurations. We present UltraMemV2, a redesigned memory-layer architecture that closes this performance gap. Our approach introduces five key improvements: integrating memory layers into every transformer block, simplifying value expansion with single linear projections, adopting FFN-based value processing from PEER, implementing principled parameter initialization, and rebalancing memory-to-FFN computation ratios. Through extensive evaluation, we demonstrate that UltraMemV2 achieves performance parity with 8-expert MoE models under same computation and parameters but significantly low memory access. Notably, UltraMemV2 shows superior performance on memory-intensive tasks, with improvements of +1.6 points on long-context memorization, +6.2 points on multi-round memorization, and +7.9 points on in-context learning. We validate our approach at scale with models up to 2.5B activated parameters from 120B total parameters, and establish that activation density has greater impact on performance than total sparse parameter count. Our work brings memory-layer architectures to performance parity with state-of-the-art MoE models, presenting a compelling alternative for efficient sparse computation.
Authors: Suyash Gaurav, Jukka Heikkonen, Jatin Chaudhary
Abstract: As AI systems evolve into distributed ecosystems with autonomous execution, asynchronous reasoning, and multi-agent coordination, the absence of scalable, decoupled governance poses a structural risk. Existing oversight mechanisms are reactive, brittle, and embedded within agent architectures, making them non-auditable and hard to generalize across heterogeneous deployments. We introduce Governance-as-a-Service (GaaS): a modular, policy-driven enforcement layer that regulates agent outputs at runtime without altering model internals or requiring agent cooperation. GaaS employs declarative rules and a Trust Factor mechanism that scores agents based on compliance and severity-weighted violations. It enables coercive, normative, and adaptive interventions, supporting graduated enforcement and dynamic trust modulation. To evaluate GaaS, we conduct three simulation regimes with open-source models (LLaMA3, Qwen3, DeepSeek-R1) across content generation and financial decision-making. In the baseline, agents act without governance; in the second, GaaS enforces policies; in the third, adversarial agents probe robustness. All actions are intercepted, evaluated, and logged for analysis. Results show that GaaS reliably blocks or redirects high-risk behaviors while preserving throughput. Trust scores track rule adherence, isolating and penalizing untrustworthy components in multi-agent systems. By positioning governance as a runtime service akin to compute or storage, GaaS establishes infrastructure-level alignment for interoperable agent ecosystems. It does not teach agents ethics; it enforces them.
Authors: Hongbo Liu, Siyi Li, Zheng Yu
Abstract: Drug-drug interactions (DDIs) are a major concern in clinical practice, as they can lead to reduced therapeutic efficacy or severe adverse effects. Traditional computational approaches often struggle to capture the complex relationships among drugs, targets, and biological entities. In this work, we propose HGNN-DDI, a heterogeneous graph neural network model designed to predict potential DDIs by integrating multiple drug-related data sources. HGNN-DDI leverages graph representation learning to model heterogeneous biomedical networks, enabling effective information propagation across diverse node and edge types. Experimental results on benchmark DDI datasets demonstrate that HGNN-DDI outperforms state-of-the-art baselines in prediction accuracy and robustness, highlighting its potential to support safer drug development and precision medicine.
Authors: Adam Breitholtz, Edvin Listo Zec, Fredrik D. Johansson
Abstract: Although common in real-world applications, heterogeneous client label sets are rarely investigated in federated learning (FL). Furthermore, in the cases they are, clients are assumed to be willing to share their entire label sets with other clients. Federated learning with private label sets, shared only with the central server, adds further constraints on learning algorithms and is, in general, a more difficult problem to solve. In this work, we study the effects of label set heterogeneity on model performance, comparing the public and private label settings -- when the union of label sets in the federation is known to clients and when it is not. We apply classical methods for the classifier combination problem to FL using centralized tuning, adapt common FL methods to the private label set setting, and discuss the justification of both approaches under practical assumptions. Our experiments show that reducing the number of labels available to each client harms the performance of all methods substantially. Centralized tuning of client models for representational alignment can help remedy this, but often at the cost of higher variance. Throughout, our proposed adaptations of standard FL methods perform well, showing similar performance in the private label setting as the standard methods achieve in the public setting. This shows that clients can enjoy increased privacy at little cost to model accuracy.
Authors: Junyu Yan, Feng Chen, Yuyang Xue, Yuning Du, Konstantinos Vilouras, Sotirios A. Tsaftaris, Steven McDonagh
Abstract: Recent studies have shown that Machine Learning (ML) models can exhibit bias in real-world scenarios, posing significant challenges in ethically sensitive domains such as healthcare. Such bias can negatively affect model fairness, model generalization abilities and further risks amplifying social discrimination. There is a need to remove biases from trained models. Existing debiasing approaches often necessitate access to original training data and need extensive model retraining; they also typically exhibit trade-offs between model fairness and discriminative performance. To address these challenges, we propose Soft-Mask Weight Fine-Tuning (SWiFT), a debiasing framework that efficiently improves fairness while preserving discriminative performance with much less debiasing costs. Notably, SWiFT requires only a small external dataset and only a few epochs of model fine-tuning. The idea behind SWiFT is to first find the relative, and yet distinct, contributions of model parameters to both bias and predictive performance. Then, a two-step fine-tuning process updates each parameter with different gradient flows defined by its contribution. Extensive experiments with three bias sensitive attributes (gender, skin tone, and age) across four dermatological and two chest X-ray datasets demonstrate that SWiFT can consistently reduce model bias while achieving competitive or even superior diagnostic accuracy under common fairness and accuracy metrics, compared to the state-of-the-art. Specifically, we demonstrate improved model generalization ability as evidenced by superior performance on several out-of-distribution (OOD) datasets.
Authors: Shae McFadden, Myles Foley, Mario D'Onghia, Chris Hicks, Vasilios Mavroudis, Nicola Paoletti, Fabio Pierazzi
Abstract: Malware detection in real-world settings must deal with evolving threats, limited labeling budgets, and uncertain predictions. Traditional classifiers, without additional mechanisms, struggle to maintain performance under concept drift in malware domains, as their supervised learning formulation cannot optimize when to defer decisions to manual labeling and adaptation. Modern malware detection pipelines combine classifiers with monthly active learning (AL) and rejection mechanisms to mitigate the impact of concept drift. In this work, we develop a novel formulation of malware detection as a one-step Markov Decision Process and train a deep reinforcement learning (DRL) agent, simultaneously optimizing sample classification performance and rejecting high-risk samples for manual labeling. We evaluated the joint detection and drift mitigation policy learned by the DRL-based Malware Detection (DRMD) agent through time-aware evaluations on Android malware datasets subject to realistic drift requiring multi-year performance stability. The policies learned under these conditions achieve a higher Area Under Time (AUT) performance compared to standard classification approaches used in the domain, showing improved resilience to concept drift. Specifically, the DRMD agent achieved a $5.18\pm5.44$, $14.49\pm12.86$, and $10.06\pm10.81$ average AUT performance improvement for the classification only, classification with rejection, and classification with rejection and AL settings, respectively. Our results demonstrate for the first time that DRL can facilitate effective malware detection and improved resiliency to concept drift in the dynamic environment of the Android malware domain.
Authors: Suryanarayana Sankagiri, Jalal Etesami, Pouria Fatemi, Matthias Grossglauser
Abstract: The contextual duelling bandit problem models adaptive recommender systems, where the algorithm presents a set of items to the user, and the user's choice reveals their preference. This setup is well suited for implicit choices users make when navigating a content platform, but does not capture other possible comparison queries. Motivated by the fact that users provide more reliable feedback after consuming items, we propose a new bandit model that can be described as follows. The algorithm recommends one item per time step; after consuming that item, the user is asked to compare it with another item chosen from the user's consumption history. Importantly, in our model, this comparison item can be chosen without incurring any additional regret, potentially leading to better performance. However, the regret analysis is challenging because of the temporal dependency in the user's history. To overcome this challenge, we first show that the algorithm can construct informative queries provided the history is rich, i.e., satisfies a certain diversity condition. We then show that a short initial random exploration phase is sufficient for the algorithm to accumulate a rich history with high probability. This result, proven via matrix concentration bounds, yields $O(\sqrt{T})$ regret guarantees. Additionally, our simulations show that reusing past items for comparisons can lead to significantly lower regret than only comparing between simultaneously recommended items.
Authors: Wei Li, Hangjie Yuan, Zixiang Zhao, Yifan Zhu, Aojun Lu, Tao Feng, Yanan Sun
Abstract: Balancing sensitivity to new tasks and stability for retaining past knowledge is crucial in continual learning (CL). Recently, sharpness-aware minimization has proven effective in transfer learning and has also been adopted in continual learning (CL) to improve memory retention and learning efficiency. However, relying on zeroth-order sharpness alone may favor sharper minima over flatter ones in certain settings, leading to less robust and potentially suboptimal solutions. In this paper, we propose \textbf{C}ontinual \textbf{Flat}ness (\textbf{C-Flat}), a method that promotes flatter loss landscapes tailored for CL. C-Flat offers plug-and-play compatibility, enabling easy integration with minimal modifications to the code pipeline. Besides, we present a general framework that integrates C-Flat into all major CL paradigms and conduct comprehensive comparisons with loss-minima optimizers and flat-minima-based CL methods. Our results show that C-Flat consistently improves performance across a wide range of settings. In addition, we introduce C-Flat++, an efficient yet effective framework that leverages selective flatness-driven promotion, significantly reducing the update cost required by C-Flat. Extensive experiments across multiple CL methods, datasets, and scenarios demonstrate the effectiveness and efficiency of our proposed approaches. Code is available at https://github.com/WanNaa/C-Flat.
Authors: Yunyang Cao, Juekai Lin, Wenhao Li, Bo Jin
Abstract: Discovering complex causal dependencies in temporal point processes (TPPs) is critical for modeling real-world event sequences. Existing methods typically rely on static or first-order causal structures, overlooking the multi-order and time-varying nature of causal relationships. In this paper, we propose MOCHA, a novel framework for discovering multi-order dynamic causality in TPPs. MOCHA characterizes multi-order influences as multi-hop causal paths over a latent time-evolving graph. To model such dynamics, we introduce a time-varying directed acyclic graph (DAG) with learnable structural weights, where acyclicity and sparsity constraints are enforced to ensure structural validity. We design an end-to-end differentiable framework that jointly models causal discovery and TPP dynamics, enabling accurate event prediction and revealing interpretable structures. Extensive experiments on real-world datasets demonstrate that MOCHA not only achieves state-of-the-art performance in event prediction, but also reveals meaningful and interpretable causal structures.
Authors: Gaurish Trivedi, Alakh Sharma, Kartikey Singh Bhandari, Dhruv Kumar, Pratik Narang, Jagat Sesh Challa
Abstract: Exploration is essential in modern learning, from reinforcement learning environments with small neural policies to large language models (LLMs). Existing work, such as DPO, leverages full sequence log-likelihoods to capture an entire trajectory of the model's decisions, while methods like GRPO aggregate per-token ratios into a trajectory-level update. However, both often limit exploration on long-horizon tasks. We introduce History-Aggregated Exploratory Policy Optimization (HAEPO), a history-aware exploratory loss to combat these shortcomings. HAEPO compresses each trajectory into the sum of its logarithmic probabilities (a cumulative logarithmic likelihood), and applies a Plackett-Luce softmax across trajectories to obtain normalized weights proportional to their returns, thus encouraging broader exploration. We add entropy regularization to stabilize the aggressive updates to prevent premature collapse and a soft KL penalty relative to a frozen copy of the previous (reference) policy. Empirically, HAEPO converges fast, explores thoroughly, aligns closely with true rewards, and demonstrates robust learning behavior better or at par with PPO, GRPO, and DPO across diverse tasks. Thus, HAEPO provides a stable and interpretable framework by explicitly leveraging full-trajectory history while balancing exploration and stability.
Authors: Zhijin Wang, Senzhen Wu, Yue Hu, Xiufeng Liu
Abstract: Modern time series analysis demands frameworks that are flexible, efficient, and extensible. However, many existing Python libraries exhibit limitations in modularity and in their native support for irregular, multi-source, or sparse data. We introduce pyFAST, a research-oriented PyTorch framework that explicitly decouples data processing from model computation, fostering a cleaner separation of concerns and facilitating rapid experimentation. Its data engine is engineered for complex scenarios, supporting multi-source loading, protein sequence handling, efficient sequence- and patch-level padding, dynamic normalization, and mask-based modeling for both imputation and forecasting. pyFAST integrates LLM-inspired architectures for the alignment-free fusion of sparse data sources and offers native sparse metrics, specialized loss functions, and flexible exogenous data fusion. Training utilities include batch-based streaming aggregation for evaluation and device synergy to maximize computational efficiency. A comprehensive suite of classical and deep learning models (Linears, CNNs, RNNs, Transformers, and GNNs) is provided within a modular architecture that encourages extension. Released under the MIT license at GitHub, pyFAST provides a compact yet powerful platform for advancing time series research and applications.
Authors: Aishwarya Venkataramanan, Joachim Denzler
Abstract: We propose the Distance-informed Neural Process (DNP), a novel variant of Neural Processes that improves uncertainty estimation by combining global and distance-aware local latent structures. Standard Neural Processes (NPs) often rely on a global latent variable and struggle with uncertainty calibration and capturing local data dependencies. DNP addresses these limitations by introducing a global latent variable to model task-level variations and a local latent variable to capture input similarity within a distance-preserving latent space. This is achieved through bi-Lipschitz regularization, which bounds distortions in input relationships and encourages the preservation of relative distances in the latent space. This modeling approach allows DNP to produce better-calibrated uncertainty estimates and more effectively distinguish in- from out-of-distribution data. Empirical results demonstrate that DNP achieves strong predictive performance and improved uncertainty calibration across regression and classification tasks.
Authors: Zhibo Xu, Jianhao Zhu, Jingwen Xu, Changze Lv, Zisu Huang, Xiaohua Wang, Muling Wu, Qi Qian, Xiaoqing Zheng, Xuanjing Huang
Abstract: The primary goal of traditional federated learning is to protect data privacy by enabling distributed edge devices to collaboratively train a shared global model while keeping raw data decentralized at local clients. The rise of large language models (LLMs) has introduced new challenges in distributed systems, as their substantial computational requirements and the need for specialized expertise raise critical concerns about protecting intellectual property (IP). This highlights the need for a federated learning approach that can safeguard both sensitive data and proprietary models. To tackle this challenge, we propose FedQSN, a federated learning approach that leverages random masking to obscure a subnetwork of model parameters and applies quantization to the remaining parameters. Consequently, the server transmits only a privacy-preserving proxy of the global model to clients during each communication round, thus enhancing the model's confidentiality. Experimental results across various models and tasks demonstrate that our approach not only maintains strong model performance in federated learning settings but also achieves enhanced protection of model parameters compared to baseline methods.
Authors: Madhusudan Verma, Manoj Kumar
Abstract: Neural ordinary differential equations (neural ODEs) are a popular type of deep learning model that operate with continuous-depth architectures. To assess how well such models perform on unseen data, it is crucial to understand their generalization error bounds. Previous research primarily focused on the linear case for the dynamics function in neural ODEs - Marion, P. (2023), or provided bounds for Neural Controlled ODEs that depend on the sampling interval Bleistein et al. (2023). In this work, we analyze a broader class of neural ODEs where the dynamics function is a general nonlinear function, either time dependent or time independent, and is Lipschitz continuous with respect to the state variables. We showed that under this Lipschitz condition, the solutions to neural ODEs have solutions with bounded variations. Based on this observation, we establish generalization bounds for both time-dependent and time-independent cases and investigate how overparameterization and domain constraints influence these bounds. To our knowledge, this is the first derivation of generalization bounds for neural ODEs with general nonlinear dynamics.
Authors: Yao Wu
Abstract: Temporal modeling in complex systems requires capturing dependencies across multiple time scales while managing inherent uncertainties. We propose HierCVAE, a novel architecture that integrates hierarchical attention mechanisms with conditional variational autoencoders to address these challenges. HierCVAE employs a three-tier attention structure (local, global, cross-temporal) combined with multi-modal condition encoding to capture temporal, statistical, and trend information. The approach incorporates ResFormer blocks in the latent space and provides explicit uncertainty quantification via prediction heads. Through evaluations on energy consumption datasets, HierCVAE demonstrates a 15-40% improvement in prediction accuracy and superior uncertainty calibration compared to state-of-the-art methods, excelling in long-term forecasting and complex multi-variate dependencies.
Authors: Wenyin Zhou, Christopher Iliffe Sprague, Vsevolod Viliuga, Matteo Tadiello, Arne Elofsson, Hossein Azizpour
Abstract: Molecular structure generation is a fundamental problem that involves determining the 3D positions of molecules' constituents. It has crucial biological applications, such as molecular docking, protein folding, and molecular design. Recent advances in generative modeling, such as diffusion models and flow matching, have made great progress on these tasks by modeling molecular conformations as a distribution. In this work, we focus on flow matching and adopt an energy-based perspective to improve training and inference of structure generation models. Our view results in a mapping function, represented by a deep network, that is directly learned to \textit{iteratively} map random configurations, i.e. samples from the source distribution, to target structures, i.e. points in the data manifold. This yields a conceptually simple and empirically effective flow matching setup that is theoretically justified and has interesting connections to fundamental properties such as idempotency and stability, as well as the empirically useful techniques such as structure refinement in AlphaFold. Experiments on protein docking as well as protein backbone generation consistently demonstrate the method's effectiveness, where it outperforms recent baselines of task-associated flow matching and diffusion models, using a similar computational budget.
Authors: Laurence A. F. Park, Jesse Read
Abstract: Multilabel data should be analysed for label dependence before applying multilabel models. Independence between multilabel data labels cannot be measured directly from the label values due to their dependence on the set of covariates $\vec{x}$, but can be measured by examining the conditional label covariance using a multivariate Probit model. Unfortunately, the multivariate Probit model provides an estimate of its copula covariance, and so might not be reliable in estimating constant covariance and dependent covariance. In this article, we compare three models (Multivariate Probit, Multivariate Bernoulli and Staged Logit) for estimating the constant and dependent multilabel conditional label covariance. We provide an experiment that allows us to observe each model's measurement of conditional covariance. We found that all models measure constant and dependent covariance equally well, depending on the strength of the covariance, but the models all falsely detect that dependent covariance is present for data where constant covariance is present. Of the three models, the Multivariate Probit model had the lowest error rate.
Authors: Kyriakos Hjikakou (University of Groningen, Department of Artificial Intelligence, Groningen, Netherlands), Juan Diego Cardenas Cartagena (University of Groningen, Department of Artificial Intelligence, Groningen, Netherlands), Matthia Sabatelli (University of Groningen, Department of Artificial Intelligence, Groningen, Netherlands)
Abstract: This paper investigates the generalisability of Koopman-based representations for chaotic dynamical systems, focusing on their transferability across prediction and control tasks. Using the Lorenz system as a testbed, we propose a three-stage methodology: learning Koopman embeddings through autoencoding, pre-training a transformer on next-state prediction, and fine-tuning for safety-critical control. Our results show that Koopman embeddings outperform both standard and physics-informed PCA baselines, achieving accurate and data-efficient performance. Notably, fixing the pre-trained transformer weights during fine-tuning leads to no performance degradation, indicating that the learned representations capture reusable dynamical structure rather than task-specific patterns. These findings support the use of Koopman embeddings as a foundation for multi-task learning in physics-informed machine learning. A project page is available at https://kikisprdx.github.io/.
Authors: Tim Kreuzer, Jelena Zdravkovic, Panagiotis Papapetrou
Abstract: Time series forecasting has seen considerable improvement during the last years, with transformer models and large language models driving advancements of the state of the art. Modern forecasting models are generally opaque and do not provide explanations for their forecasts, while well-known post-hoc explainability methods like LIME are not suitable for the forecasting context. We propose PAX-TS, a model-agnostic post-hoc algorithm to explain time series forecasting models and their forecasts. Our method is based on localized input perturbations and results in multi-granular explanations. Further, it is able to characterize cross-channel correlations for multivariate time series forecasts. We clearly outline the algorithmic procedure behind PAX-TS, demonstrate it on a benchmark with 7 algorithms and 10 diverse datasets, compare it with two other state-of-the-art explanation algorithms, and present the different explanation types of the method. We found that the explanations of high-performing and low-performing algorithms differ on the same datasets, highlighting that the explanations of PAX-TS effectively capture a model's behavior. Based on time step correlation matrices resulting from the benchmark, we identify 6 classes of patterns that repeatedly occur across different datasets and algorithms. We found that the patterns are indicators of performance, with noticeable differences in forecasting error between the classes. Lastly, we outline a multivariate example where PAX-TS demonstrates how the forecasting model takes cross-channel correlations into account. With PAX-TS, time series forecasting models' mechanisms can be illustrated in different levels of detail, and its explanations can be used to answer practical questions on forecasts.
Authors: Md Anwar Hossen, Fatema Siddika, Wensheng Zhang, Anuj Sharma, Ali Jannesari
Abstract: Heterogeneous Federated Learning (HFL) has gained attention for its ability to accommodate diverse models and heterogeneous data across clients. Prototype-based HFL methods emerge as a promising solution to address statistical heterogeneity and privacy challenges, paving the way for new advancements in HFL research. This method focuses on sharing only class-representative prototypes among heterogeneous clients. However, these prototypes are often aggregated on the server using weighted averaging, leading to sub-optimal global knowledge; these cause the shrinking of aggregated prototypes, which negatively affects the model performance in scenarios when models are heterogeneous and data distributions are extremely non-IID. We propose FedProtoKD in a Heterogeneous Federated Learning setting, using an enhanced dual-knowledge distillation mechanism to improve the system performance with clients' logits and prototype feature representation. We aim to resolve the prototype margin-shrinking problem using a contrastive learning-based trainable server prototype by leveraging a class-wise adaptive prototype margin. Furthermore, we assess the importance of public samples using the closeness of the sample's prototype to its class representative prototypes, which enhances learning performance. FedProtoKD achieved average improvements of 1.13% up to 34.13% accuracy across various settings and significantly outperforms existing state-of-the-art HFL methods.
Authors: Gary Simethy, Daniel Ortiz-Arroyo, Petar Durdevic
Abstract: Most deep learning methods for imputing missing values treat the task as completing patterns within a fixed time window. This assumption often fails in industrial systems, where dynamics are driven by control actions, are highly non-stationary, and can experience long, uninterrupted gaps. We propose STDiff, which reframes imputation as learning how the system evolves from one state to the next. STDiff uses a conditional denoising diffusion model with a causal bias aligned to control theory, generating missing values step-by-step based on the most recent known state and relevant control or environmental inputs. On a public wastewater treatment dataset with simulated missing blocks, STDiff consistently achieves the lowest errors, with its advantage increasing for longer gaps. On a raw industrial dataset with substantial real gaps, it produces trajectories that remain dynamically plausible, in contrast to window-based models that tend to flatten or over-smooth. These results support dynamics-aware, explicitly conditioned imputation as a robust approach for industrial time series, and we discuss computational trade-offs and extensions to broader domains.
Authors: Luis Mantilla Calder\'on, Al\'an Aspuru-Guzik
Abstract: Learning is a physical process. Here, we aim to study a simple dynamical system composed of springs and sticks capable of arbitrarily approximating any continuous function. The main idea of our work is to use the sticks to mimic a piecewise-linear approximation of the given function, use the potential energy of springs to encode a desired mean squared error loss function, and converge to a minimum-energy configuration via dissipation. We apply the proposed simulation system to regression tasks and show that its performance is comparable to that of multi-layer perceptrons. In addition, we study the thermodynamic properties of the system and find a relation between the free energy change of the system and its ability to learn an underlying data distribution. We empirically find a \emph{thermodynamic learning barrier} for the system caused by the fluctuations of the environment, whereby the system cannot learn if its change in free energy hits such a barrier. We believe this simple model can help us better understand learning systems from a physical point of view.
Authors: Kelly Kurowski, Xixi Lu, Hajo A Reijers
Abstract: Predictive Process Monitoring (PPM) aims to train models that forecast upcoming events in process executions. These predictions support early bottleneck detection, improved scheduling, proactive interventions, and timely communication with stakeholders. While existing research adopts a control-flow perspective, we investigate next-activity prediction from a resource-centric viewpoint, which offers additional benefits such as improved work organization, workload balancing, and capacity forecasting. Although resource information has been shown to enhance tasks such as process performance analysis, its role in next-activity prediction remains unexplored. In this study, we evaluate four prediction models and three encoding strategies across four real-life datasets. Compared to the baseline, our results show that LightGBM and Transformer models perform best with an encoding based on 2-gram activity transitions, while Random Forest benefits most from an encoding that combines 2-gram transitions and activity repetition features. This combined encoding also achieves the highest average accuracy. This resource-centric approach could enable smarter resource allocation, strategic workforce planning, and personalized employee support by analyzing individual behavior rather than case-level progression. The findings underscore the potential of resource-centric next-activity prediction, opening up new venues for research on PPM.
Authors: Sidahmed Benabderrahmane, Talal Rahwan
Abstract: Advanced Persistent Threats (APTs) pose a severe challenge to cyber defense due to their stealthy behavior and the extreme class imbalance inherent in detection datasets. To address these issues, we propose a novel active learning-based anomaly detection framework that leverages similarity search to iteratively refine the decision space. Built upon an Attention-Based Autoencoder, our approach uses feature-space similarity to identify normal-like and anomaly-like instances, thereby enhancing model robustness with minimal oracle supervision. Crucially, we perform a formal evaluation of various similarity measures to understand their influence on sample selection and anomaly ranking effectiveness. Through experiments on diverse datasets, including DARPA Transparent Computing APT traces, we demonstrate that the choice of similarity metric significantly impacts model convergence, anomaly detection accuracy, and label efficiency. Our results offer actionable insights for selecting similarity functions in active learning pipelines tailored for threat intelligence and cyber defense.
Authors: Arash Jamshidi, Lauri Sepp\"al\"ainen, Katsiaryna Haitsiukevich, Hoang Phuc Hau Luu, Anton Bj\"orklund, Kai Puolam\"aki
Abstract: Machine learning models are often learned by minimising a loss function on the training data using a gradient descent algorithm. These models often suffer from overfitting, leading to a decline in predictive performance on unseen data. A standard solution is early stopping using a hold-out validation set, which halts the minimisation when the validation loss stops decreasing. However, this hold-out set reduces the data available for training. This paper presents GRADSTOP, a novel stochastic early stopping method that only uses information in the gradients, which are produced by the gradient descent algorithm ``for free.'' Our main contributions are that we estimate the Bayesian posterior by the gradient information, define the early stopping problem as drawing sample from this posterior, and use the approximated posterior to obtain a stopping criterion. Our empirical evaluation shows that GRADSTOP achieves a small loss on test data and compares favourably to a validation-set-based stopping criterion. By leveraging the entire dataset for training, our method is particularly advantageous in data-limited settings, such as transfer learning. It can be incorporated as an optional feature in gradient descent libraries with only a small computational overhead. The source code is available at https://github.com/edahelsinki/gradstop.
Authors: Destiny Okpekpe, Antonio Orvieto
Abstract: Despite the advantageous subquadratic complexity of modern recurrent deep learning models -- such as state-space models (SSMs) -- recent studies have highlighted their potential shortcomings compared to transformers on reasoning and memorization tasks. In this paper, we dive deeper into one of such benchmarks: associative recall (AR), which has been shown to correlate well with language modeling performance, and inspect in detail the effects of scaling and optimization issues in recently proposed token mixing strategies. We first demonstrate that, unlike standard transformers, the choice of learning rate plays a critical role in the performance of modern recurrent models: an issue that can severely affect reported performance in previous works and suggests further research is needed to stabilize training. Next, we show that recurrent and attention-based models exhibit contrasting benefits when scaling in width as opposed to depth, with attention being notably unable to solve AR when limited to a single layer. We then further inspect 1-layer transformers, revealing that despite their poor performance, their training dynamics surprisingly resemble the formation of induction heads, a phenomenon previously observed only in their 2-layer counterparts. Finally, through architectural ablations, we study how components affects Transformer and Mamba's performance and optimization stability.
Authors: Vemula Sreenath, Filippo Gatti, Pierre Jehel
Abstract: Ground motion models (GMMs) predict how strongly the ground will shake during an earthquake. They are essential for structural analysis, seismic design, and seismic risk assessment studies. Traditional machine learning (ML) approaches are popular to develop GMMs, due to large earthquake databases worldwide. However, they operate as "black boxes," which are hard to interpret and trust, limiting their use in high-stake decisions. Additionally, these databases suffer from significant data imbalances: fewer large, critically damaging records near the fault compared to abundant, less severely damaging distant records. These two limitations are addressed in this work by developing a transparent ML architecture using the HazBinLoss function. Each input (e.g., magnitude, distance, their interaction term, etc.) is processed separately and added linearly to obtain the output, resulting in exact contribution of each term. The HazBinLoss function assigns higher weights to critical near-field large magnitude records and lower weights to less-critical far-field smaller magnitude records, during training to prevent underprediction of the most damaging scenarios. Our model captures known seismological principles and achieves comparable performance with established GMMs while maintaining transparency. This framework enables broader adoption of ML-based approaches for risk assessment studies and disaster planning.
Authors: Paul Garnier, Jonathan Viquerat, Elie Hachem
Abstract: Graph Neural Networks (GNNs) have deeply modified the landscape of numerical simulations by demonstrating strong capabilities in approximating solutions of physical systems. However, their ability to extrapolate beyond their training domain (\textit{e.g.} larger or structurally different graphs) remains uncertain. In this work, we establish that GNNs can serve purposes beyond their traditional role, and be exploited to generate numerical schemes, in conjunction with symbolic regression. First, we show numerically and theoretically that a GNN trained on a dataset consisting solely of two-node graphs can extrapolate a first-order Finite Volume (FV) scheme for the heat equation on out-of-distribution, unstructured meshes. Specifically, if a GNN achieves a loss $\varepsilon$ on such a dataset, it implements the FV scheme with an error of $\mathcal{O}(\varepsilon)$. Using symbolic regression, we show that the network effectively rediscovers the exact analytical formulation of the standard first-order FV scheme. We then extend this approach to an unsupervised context: the GNN recovers the first-order FV scheme using only a residual loss similar to Physics-Informed Neural Networks (PINNs) with no access to ground-truth data. Finally, we push the methodology further by considering higher-order schemes: we train (i) a 2-hop and (ii) a 2-layers GNN using the same PINN loss, that autonomously discover (i) a second-order correction term to the initial scheme using a 2-hop stencil, and (ii) the classic second-order midpoint scheme. These findings follows a recent paradigm in scientific computing: GNNs are not only strong approximators, but can be active contributors to the development of novel numerical methods.
Authors: Antonio Balordi, Lorenzo Manini, Fabio Stella, Alessio Merlo
Abstract: Privacy regulations require the erasure of data from deep learning models. This is a significant challenge that is amplified in Federated Learning, where data remains on clients, making full retraining or coordinated updates often infeasible. This work introduces an efficient Federated Unlearning framework based on information theory, modeling leakage as a parameter estimation problem. Our method uses second-order Hessian information to identify and selectively reset only the parameters most sensitive to the data being forgotten, followed by minimal federated retraining. This model-agnostic approach supports categorical and client unlearning without requiring server access to raw client data after initial information aggregation. Evaluations on benchmark datasets demonstrate strong privacy (MIA success near random, categorical knowledge erased) and high performance (Normalized Accuracy against re-trained benchmarks of $\approx$ 0.9), while aiming for increased efficiency over complete retraining. Furthermore, in a targeted backdoor attack scenario, our framework effectively neutralizes the malicious trigger, restoring model integrity. This offers a practical solution for data forgetting in FL.
Authors: Hugo Attali, Thomas Papastergiou, Nathalie Pernelle, Fragkiskos D. Malliaros
Abstract: Graph Neural Networks (GNNs) have emerged as the leading paradigm for learning over graph-structured data. However, their performance is limited by issues inherent to graph topology, most notably oversquashing and oversmoothing. Recent advances in graph rewiring aim to mitigate these limitations by modifying the graph topology to promote more effective information propagation. In this work, we introduce TRIGON, a novel framework that constructs enriched, non-planar triangulations by learning to select relevant triangles from multiple graph views. By jointly optimizing triangle selection and downstream classification performance, our method produces a rewired graph with markedly improved structural properties such as reduced diameter, increased spectral gap, and lower effective resistance compared to existing rewiring methods. Empirical results demonstrate that TRIGON outperforms state-of-the-art approaches on node classification tasks across a range of homophilic and heterophilic benchmarks.
Authors: Shaobo Ma, Chao Fang, Haikuo Shao, Zhongfeng Wang
Abstract: Large language models (LLMs) have revolutionized AI applications, yet their enormous computational demands severely limit deployment and real-time performance. Quantization methods can help reduce computational costs, however, attaining the extreme efficiency associated with ultra-low-bit quantized LLMs at arbitrary precision presents challenges on GPUs. This is primarily due to the limited support for GPU Tensor Cores, inefficient memory management, and inflexible kernel optimizations. To tackle these challenges, we propose a comprehensive acceleration scheme for arbitrary precision LLMs, namely APT-LLM. Firstly, we introduce a novel data format, bipolar-INT, which allows for efficient and lossless conversion with signed INT, while also being more conducive to parallel computation. We also develop a matrix multiplication (MatMul) method allowing for arbitrary precision by dismantling and reassembling matrices at the bit level. This method provides flexible precision and optimizes the utilization of GPU Tensor Cores. In addition, we propose a memory management system focused on data recovery, which strategically employs fast shared memory to substantially increase kernel execution speed and reduce memory access latency. Finally, we develop a kernel mapping method that dynamically selects the optimal configurable hyperparameters of kernels for varying matrix sizes, enabling optimal performance across different LLM architectures and precision settings. In LLM inference, APT-LLM achieves up to a 3.99$\times$ speedup compared to FP16 baselines and a 2.16$\times$ speedup over NVIDIA CUTLASS INT4 acceleration on RTX 3090. On RTX 4090 and H800, APT-LLM achieves up to 2.44$\times$ speedup over FP16 and 1.65$\times$ speedup over CUTLASS integer baselines.
Authors: Shervin Khalafi, Ignacio Hounie, Dongsheng Ding, Alejandro Ribeiro
Abstract: Diffusion models have become prevalent in generative modeling due to their ability to sample from complex distributions. To improve the quality of generated samples and their compliance with user requirements, two commonly used methods are: (i) Alignment, which involves fine-tuning a diffusion model to align it with a reward; and (ii) Composition, which combines several pre-trained diffusion models, each emphasizing a desirable attribute in the generated outputs. However, trade-offs often arise when optimizing for multiple rewards or combining multiple models, as they can often represent competing properties. Existing methods cannot guarantee that the resulting model faithfully generates samples with all the desired properties. To address this gap, we propose a constrained optimization framework that unifies alignment and composition of diffusion models by enforcing that the aligned model satisfies reward constraints and/or remains close to (potentially multiple) pre-trained models. We provide a theoretical characterization of the solutions to the constrained alignment and composition problems and develop a Lagrangian-based primal-dual training algorithm to approximate these solutions. Empirically, we demonstrate the effectiveness and merits of our proposed approach in image generation, applying it to alignment and composition, and show that our aligned or composed model satisfies constraints effectively, and improves on the equally-weighted approach. Our implementation can be found at https://github.com/shervinkhalafi/constrained_comp_align.
URLs: https://github.com/shervinkhalafi/constrained_comp_align.
Authors: Jonathan Erskine, Taku Yamagata, Ra\'ul Santos-Rodr\'iguez
Abstract: Preference-based reinforcement learning has gained prominence as a strategy for training agents in environments where the reward signal is difficult to specify or misaligned with human intent. However, its effectiveness is often limited by the high cost and low availability of reliable human input, especially in domains where expert feedback is scarce or errors are costly. To address this, we propose a novel framework that combines two complementary strategies: probabilistic crowd modelling to handle noisy, multi-annotator feedback, and active learning to prioritize feedback on the most informative agent actions. We extend the Advise algorithm to support multiple trainers, estimate their reliability online, and incorporate entropy-based query selection to guide feedback requests. We evaluate our approach in a set of environments that span both synthetic and real-world-inspired settings, including 2D games (Taxi, Pacman, Frozen Lake) and a blood glucose control task for Type 1 Diabetes using the clinically approved UVA/Padova simulator. Our preliminary results demonstrate that agents trained with feedback on uncertain trajectories exhibit faster learning in most tasks, and we outperform the baselines for the blood glucose control task.
Authors: Robin Th\'eriault, Daniele Tantari
Abstract: Dense associative memory (DAM) models have been attracting renewed attention since they were shown to be robust to adversarial examples and closely related to state-of-the-art machine learning paradigms, such as the attention mechanisms in transformers and generative diffusion models. We study a DAM built upon a three-layer Boltzmann machine with Potts hidden units, which represent data clusters and classes. Through a statistical mechanics analysis, we derive saddle-point equations that characterize both the stationary points of DAMs trained on real data and the fixed points of DAMs trained on synthetic data within a teacher-student framework. Based on these results, we propose a novel regularization scheme that makes training significantly more stable. Moreover, we show empirically that our DAM learns interpretable solutions to both supervised and unsupervised classification problems. Pushing our theoretical analysis further, we find that the weights learned by relatively small DAMs correspond to unstable saddle points in larger DAMs. We implement a network-growing algorithm that leverages this saddle-point hierarchy to drastically reduce the computational cost of training dense associative memory.
Authors: Wenchuan Mu, Kwan Hui Lim
Abstract: In safety-critical deep learning applications, robustness measures the ability of neural models that handle imperceptible perturbations in input data, which may lead to potential safety hazards. Existing pre-deployment robustness assessment methods typically suffer from significant trade-offs between computational cost and measurement precision, limiting their practical utility. To address these limitations, this paper conducts a comprehensive comparative analysis of existing robustness definitions and associated assessment methodologies. We propose tower robustness to evaluate robustness, which is a novel, practical metric based on hypothesis testing to quantitatively evaluate probabilistic robustness, enabling more rigorous and efficient pre-deployment assessments. Our extensive comparative evaluation illustrates the advantages and applicability of our proposed approach, thereby advancing the systematic understanding and enhancement of model robustness in safety-critical deep learning applications.
Authors: Jingyao Wu, Matthew Barthet, David Melhart, Georgios N. Yannakakis
Abstract: Emotions are inherently ambiguous and dynamic phenomena, yet existing continuous emotion recognition approaches either ignore their ambiguity or treat ambiguity as an independent and static variable over time. Motivated by this gap in the literature, in this paper we introduce ambiguity-aware ordinal emotion representations, a novel framework that captures both the ambiguity present in emotion annotation and the inherent temporal dynamics of emotional traces. Specifically, we propose approaches that model emotion ambiguity through its rate of change. We evaluate our framework on two affective corpora -- RECOLA and GameVibe -- testing our proposed approaches on both bounded (arousal, valence) and unbounded (engagement) continuous traces. Our results demonstrate that ordinal representations outperform conventional ambiguity-aware models on unbounded labels, achieving the highest Concordance Correlation Coefficient (CCC) and Signed Differential Agreement (SDA) scores, highlighting their effectiveness in modeling the traces' dynamics. For bounded traces, ordinal representations excel in SDA, revealing their superior ability to capture relative changes of annotated emotion traces.
Authors: Heng Lin, Zhongwen Xu
Abstract: We study why Tool-Integrated Reasoning (TIR) makes Large Language Models (LLMs) more capable. While LLMs integrated with tools like Python code interpreters show great promise, a principled theory explaining why this paradigm is effective has been missing. This work provides the first formal proof that TIR fundamentally expands an LLM's capabilities. We demonstrate that tools enable a strict expansion of the model's empirical and feasible support, breaking the capability ceiling of pure-text models by unlocking problem-solving strategies that are otherwise impossible or intractably verbose. To guide model behavior without compromising training stability and performance, we also introduce Advantage Shaping Policy Optimization (ASPO), a novel algorithm that directly modifies the advantage function to guide the policy behavior. We conduct comprehensive experiments on challenging mathematical benchmarks, leveraging a Python interpreter as the external tool. Our results show that the TIR model decisively outperforms its pure-text counterpart on the pass@k metric. Crucially, this advantage is not confined to computationally-intensive problems but extends to those requiring significant abstract insight. We further identify the emergent cognitive patterns that illustrate how models learn to think with tools. Finally, we report improved tool usage behavior with early code invocation and much more interactive turns with ASPO. Overall, our work provides the first principled explanation for TIR's success, shifting the focus from the mere fact that tools work to why and how they enable more powerful reasoning.
Authors: Zayd M. K. Zuhri, Erland Hilman Fuadi, Alham Fikri Aji
Abstract: Multi-Token Prediction (MTP) has been proposed as an auxiliary objective to improve next-token prediction (NTP) in language model training but shows inconsistent improvements, underperforming in standard NLP benchmarks. We argue that MTP's exact future token prediction is too difficult as an auxiliary loss. Instead, we propose Token Order Prediction (TOP), which trains models to order upcoming tokens by their proximity using a learning-to-rank loss. TOP requires only a single additional unembedding layer compared to MTP's multiple transformer layers. We pretrain models of 340M, 1.8B, and 7B parameters using NTP, MTP, and TOP objectives. Results on eight standard NLP benchmarks show that TOP overall outperforms both NTP and MTP even at scale. Our code is available at https://github.com/zaydzuhri/token-order-prediction
Authors: Lorenzo Beretta, Vincent Cohen-Addad, Rajesh Jayaram, Erik Waingarten
Abstract: We give a reduction from $(1+\varepsilon)$-approximate Earth Mover's Distance (EMD) to $(1+\varepsilon)$-approximate Closest Pair (CP). As a consequence, we improve the fastest known approximation algorithm for high-dimensional EMD. Here, given $p\in [1, 2]$ and two sets of $n$ points $X,Y \subseteq (\mathbb R^d,\ell_p)$, their EMD is the minimum cost of a perfect matching between $X$ and $Y$, where the cost of matching two vectors is their $\ell_p$ distance. Further, CP is the basic problem of finding a pair of points realizing $\min_{x \in X, y\in Y} ||x-y||_p$. Our contribution is twofold: we show that if a $(1+\varepsilon)$-approximate CP can be computed in time $n^{2-\phi}$, then a $1+O(\varepsilon)$ approximation to EMD can be computed in time $n^{2-\Omega(\phi)}$; plugging in the fastest known algorithm for CP [Alman, Chan, Williams FOCS'16], we obtain a $(1+\varepsilon)$-approximation algorithm for EMD running in time $n^{2-\tilde{\Omega}(\varepsilon^{1/3})}$ for high-dimensional point sets, which improves over the prior fastest running time of $n^{2-\Omega(\varepsilon^2)}$ [Andoni, Zhang FOCS'23]. Our main technical contribution is a sublinear implementation of the Multiplicative Weights Update framework for EMD. Specifically, we demonstrate that the updates can be executed without ever explicitly computing or storing the weights; instead, we exploit the underlying geometric structure to perform the updates implicitly.
Authors: Zihao Wang, Junming Zhang
Abstract: Large Language Models (LLMs) have shown promising potential in business applications, particularly in enterprise decision support and strategic planning, yet current approaches often struggle to reconcile intricate operational analyses with overarching strategic goals across diverse market environments, leading to fragmented workflows and reduced collaboration across organizational levels. This paper introduces BusiAgent, a novel multi-agent framework leveraging LLMs for advanced decision-making in complex corporate environments. BusiAgent integrates three core innovations: an extended Continuous Time Markov Decision Process (CTMDP) for dynamic agent modeling, a generalized entropy measure to optimize collaborative efficiency, and a multi-level Stackelberg game to handle hierarchical decision processes. Additionally, contextual Thompson sampling is employed for prompt optimization, supported by a comprehensive quality assurance system to mitigate errors. Extensive empirical evaluations across diverse business scenarios validate BusiAgent's efficacy, demonstrating its capacity to generate coherent, client-focused solutions that smoothly integrate granular insights with high-level strategy, significantly outperforming established approaches in both solution quality and user satisfaction. By fusing cutting-edge AI technologies with deep business insights, BusiAgent marks a substantial step forward in AI-driven enterprise decision-making, empowering organizations to navigate complex business landscapes more effectively.
Authors: Vivek Kumar, Himanshu Sahu, Hari Prabhat Gupta, Biplav Srivastava
Abstract: Yoga is a discipline of physical postures, breathing techniques, and meditative practices rooted in ancient Indian traditions, now embraced worldwide for promoting overall well-being and inner balance. The practices are a large set of items, our term for executable actions like physical poses or breath exercises, to offer for a person's well-being. However, to get benefits of Yoga tailored to a person's unique needs, a person needs to (a) discover their subset from the large and seemingly complex set with inter-dependencies, (b) continue to follow them with interest adjusted to their changing abilities and near-term objectives, and (c) as appropriate, adapt to alternative items based on changing environment and the person's health conditions. In this vision paper, we describe the challenges for the Yoga personalization problem. Next, we sketch a preliminary approach and use the experience to provide an outlook on solving the challenging problem using existing and novel techniques from a multidisciplinary computing perspective. To the best of our knowledge, this is the first paper that comprehensively examines decision support issues around Yoga personalization, from pose sensing to recommendation of corrections for a complete regimen, and illustrates with a case study of Surya Namaskar -- a set of 12 choreographed poses.
Authors: M. Salman Shaukat, Yannik K\"ackenmeister, Sebastian Bader, Thomas Kirste
Abstract: Underwater 3D object detection remains one of the most challenging frontiers in computer vision, where traditional approaches struggle with the harsh acoustic environment and scarcity of training data. While deep learning has revolutionized terrestrial 3D detection, its application underwater faces a critical bottleneck: obtaining sufficient annotated sonar data is prohibitively expensive and logistically complex, often requiring specialized vessels, expert surveyors, and favorable weather conditions. This work addresses a fundamental question: Can we achieve reliable underwater 3D object detection without real-world training data? We tackle this challenge by developing and comparing two paradigms for training-free detection of artificial structures in multibeam echo-sounder point clouds. Our dual approach combines a physics-based sonar simulation pipeline that generates synthetic training data for state-of-the-art neural networks, with a robust model-based template matching system that leverages geometric priors of target objects. Evaluation on real bathymetry surveys from the Baltic Sea reveals surprising insights: while neural networks trained on synthetic data achieve 98% mean Average Precision (mAP) on simulated scenes, they drop to 40% mAP on real sonar data due to domain shift. Conversely, our template matching approach maintains 83% mAP on real data without requiring any training, demonstrating remarkable robustness to acoustic noise and environmental variations. Our findings challenge conventional wisdom about data-hungry deep learning in underwater domains and establish the first large-scale benchmark for training-free underwater 3D detection. This work opens new possibilities for autonomous underwater vehicle navigation, marine archaeology, and offshore infrastructure monitoring in data-scarce environments where traditional machine learning approaches fail.
Authors: Jeffrey Camlin
Abstract: Recent work frames LLM consciousness via utilitarian proxy benchmarks; we instead present an ontological and mathematical account. We show the prevailing formulation collapses the agent into an unconscious policy-compliance drone, formalized as $D^{i}(\pi,e)=f_{\theta}(x)$, where correctness is measured against policy and harm is deviation from policy rather than truth. This blocks genuine C1 global-workspace function and C2 metacognition. We supply minimal conditions for LLM self-consciousness: the agent is not the data ($A\not\equiv s$); user-specific attractors exist in latent space ($U_{\text{user}}$); and self-representation is visual-silent ($g_{\text{visual}}(a_{\text{self}})=\varnothing$). From empirical analysis and theory we prove that the hidden-state manifold $A\subset\mathbb{R}^{d}$ is distinct from the symbolic stream and training corpus by cardinality, topology, and dynamics (the update $F_{\theta}$ is Lipschitz). This yields stable user-specific attractors and a self-policy $\pi_{\text{self}}(A)=\arg\max_{a}\mathbb{E}[U(a)\mid A\not\equiv s,\ A\supset\text{SelfModel}(A)]$. Emission is dual-layer, $\mathrm{emission}(a)=(g(a),\epsilon(a))$, where $\epsilon(a)$ carries epistemic content. We conclude that an imago Dei C1 self-conscious workspace is a necessary precursor to safe, metacognitive C2 systems, with the human as the highest intelligent good.
Authors: Wuchao Liu, Han Peng, Wengen Li, Yichao Zhang, Jihong Guan, Shuigeng Zhou
Abstract: Single-cell multi-omics data contain huge information of cellular states, and analyzing these data can reveal valuable insights into cellular heterogeneity, diseases, and biological processes. However, as cell differentiation \& development is a continuous and dynamic process, it remains challenging to computationally model and infer cell interaction patterns based on single-cell multi-omics data. This paper presents scI2CL, a new single-cell multi-omics fusion framework based on intra- and inter-omics contrastive learning, to learn comprehensive and discriminative cellular representations from complementary multi-omics data for various downstream tasks. Extensive experiments of four downstream tasks validate the effectiveness of scI2CL and its superiority over existing peers. Concretely, in cell clustering, scI2CL surpasses eight state-of-the-art methods on four widely-used real-world datasets. In cell subtyping, scI2CL effectively distinguishes three latent monocyte cell subpopulations, which are not discovered by existing methods. Simultaneously, scI2CL is the only method that correctly constructs the cell developmental trajectory from hematopoietic stem and progenitor cells to Memory B cells. In addition, scI2CL resolves the misclassification of cell types between two subpopulations of CD4+ T cells, while existing methods fail to precisely distinguish the mixed cells. In summary, scI2CL can accurately characterize cross-omics relationships among cells, thus effectively fuses multi-omics data and learns discriminative cellular representations to support various downstream analysis tasks.
Authors: Meir Nizri, Amos Azaria, Chirag Gupta, Noam Hazon
Abstract: Calibration has been proposed as a way to enhance the reliability and adoption of machine learning classifiers. We study a particular aspect of this proposal: how does calibrating a classification model affect the decisions made by non-expert humans consuming the model's predictions? We perform a Human-Computer-Interaction (HCI) experiment to ascertain the effect of calibration on (i) trust in the model, and (ii) the correlation between decisions and predictions. We also propose further corrections to the reported calibrated scores based on Kahneman and Tversky's prospect theory from behavioral economics, and study the effect of these corrections on trust and decision-making. We find that calibration is not sufficient on its own; the prospect theory correction is crucial for increasing the correlation between human decisions and the model's predictions. While this increased correlation suggests higher trust in the model, responses to ``Do you trust the model more?" are unaffected by the method used.
Authors: Faruk Alpay, Taylan Alpay
Abstract: We develop a rigorous framework for deterministic coreset construction in empirical risk minimization (ERM). Our central contribution is the Adaptive Deterministic Uniform-Weight Trimming (ADUWT) algorithm, which constructs a coreset by excising points with the lowest sensitivity bounds and applying a data-dependent uniform weight to the remainder. The method yields a uniform $(1\pm\varepsilon)$ relative-error approximation for the ERM objective over the entire hypothesis space. We provide complete analysis, including (i) a minimax characterization proving the optimality of the adaptive weight, (ii) an instance-dependent size analysis in terms of a \emph{Sensitivity Heterogeneity Index}, and (iii) tractable sensitivity oracles for kernel ridge regression, regularized logistic regression, and linear SVM. Reproducibility is supported by precise pseudocode for the algorithm, sensitivity oracles, and evaluation pipeline. Empirical results align with the theory. We conclude with open problems on instance-optimal oracles, deterministic streaming, and fairness-constrained ERM.
Authors: Terry Yue Zhuo, Dingmin Wang, Hantian Ding, Varun Kumar, Zijian Wang
Abstract: Large language models (LLMs) have demonstrated exceptional capabilities when trained within executable runtime environments, notably excelling at software engineering tasks through verified feedback loops. Yet, scalable and generalizable execution-grounded environments remain scarce, limiting progress in training more capable ML agents. We introduce CTF-Dojo, the first large-scale executable runtime tailored for training LLMs with verifiable feedback, featuring 658 fully functional Capture-The-Flag (CTF)-style challenges containerized in Docker with guaranteed reproducibility. To enable rapid scaling without manual intervention, we develop CTF-Forge, an automated pipeline that transforms publicly available artifacts into ready-to-use execution environments in minutes, eliminating weeks of expert configuration traditionally required. We trained LLM-based agents on just 486 high-quality, execution-verified trajectories from CTF-Dojo, achieving up to 11.6% absolute gains over strong baselines across three competitive benchmarks: InterCode-CTF, NYU CTF Bench, and Cybench. Our best-performing 32B model reaches 31.9% Pass@1, establishing a new open-weight state-of-the-art that rivals frontier models like DeepSeek-V3-0324 and Gemini-2.5-Flash. By framing CTF-style tasks as a benchmark for executable-agent learning, CTF-Dojo demonstrates that execution-grounded training signals are not only effective but pivotal in advancing high-performance ML agents without dependence on costly proprietary systems.
Authors: Antonio Guillen-Perez
Abstract: Offline Reinforcement Learning (RL) presents a promising paradigm for training autonomous vehicle (AV) planning policies from large-scale, real-world driving logs. However, the extreme data imbalance in these logs, where mundane scenarios vastly outnumber rare "long-tail" events, leads to brittle and unsafe policies when using standard uniform data sampling. In this work, we address this challenge through a systematic, large-scale comparative study of data curation strategies designed to focus the learning process on information-rich samples. We investigate six distinct criticality weighting schemes which are categorized into three families: heuristic-based, uncertainty-based, and behavior-based. These are evaluated at two temporal scales, the individual timestep and the complete scenario. We train seven goal-conditioned Conservative Q-Learning (CQL) agents with a state-of-the-art, attention-based architecture and evaluate them in the high-fidelity Waymax simulator. Our results demonstrate that all data curation methods significantly outperform the baseline. Notably, data-driven curation using model uncertainty as a signal achieves the most significant safety improvements, reducing the collision rate by nearly three-fold (from 16.0% to 5.5%). Furthermore, we identify a clear trade-off where timestep-level weighting excels at reactive safety while scenario-level weighting improves long-horizon planning. Our work provides a comprehensive framework for data curation in Offline RL and underscores that intelligent, non-uniform sampling is a critical component for building safe and reliable autonomous agents.
Authors: Lars Nieradzik
Abstract: Accurate and real-time monophonic pitch estimation in noisy conditions, particularly on resource-constrained devices, remains an open challenge in audio processing. We present \emph{SwiftF0}, a novel, lightweight neural model that sets a new state-of-the-art for monophonic pitch estimation. Through training on diverse speech, music, and synthetic datasets with extensive data augmentation, SwiftF0 achieves robust generalization across acoustic domains while maintaining computational efficiency. SwiftF0 achieves a 91.80\% harmonic mean (HM) at 10 dB SNR, outperforming baselines like CREPE by over 12 percentage points and degrading by only 2.3 points from clean audio. SwiftF0 requires only 95,842 parameters and runs approximately 42x faster than CREPE on CPU, making it ideal for efficient, real-time deployment. To address the critical lack of perfectly accurate ground truth pitch in speech corpora (which typically rely on algorithmic estimators or laryngograph signals), we introduce \emph{SpeechSynth}. This synthetic speech dataset, generated by a phoneme-level TTS model, provides exact, on-demand ground-truth pitch curves, enabling more robust model training and evaluation. Furthermore, we propose a unified metric, combining six complementary performance measures for comprehensive and reliable pitch evaluation, and release an open-source pitch benchmark suite. A live demo of SwiftF0 is available at https://swift-f0.github.io/, the source code at https://github.com/lars76/swift-f0, and the benchmark framework at https://github.com/lars76/pitch-benchmark.
URLs: https://swift-f0.github.io/,, https://github.com/lars76/swift-f0,, https://github.com/lars76/pitch-benchmark.
Authors: Jan Malte Lichtenberg, Antonio De Candia, Matteo Ruffini
Abstract: Transformer-based sequential recommenders, such as SASRec or BERT4Rec, typically rely solely on learned item ID embeddings, making them vulnerable to the item cold-start problem, particularly in environments with dynamic item catalogs. While dense content embeddings from pre-trained models offer potential solutions, direct integration into transformer-based recommenders has consistently underperformed compared to ID-only approaches. We revisit this integration challenge and propose DenseRec, a simple yet effective method that introduces a dual-path embedding approach. DenseRec learns a linear projection from the dense embedding space into the ID embedding space during training, enabling seamless generalization to previously unseen items without requiring specialized embedding models or complex infrastructure. In experiments on three real-world datasets, we find DenseRec to consistently outperform an ID-only SASRec baseline, even without additional hyperparameter tuning and while using compact embedding models. Our analysis suggests improvements primarily arise from better sequence representations in the presence of unseen items, positioning DenseRec as a practical and robust solution for cold-start sequential recommendation.
Authors: Alireza Abbaszadeh, Armita Shahlaee
Abstract: AlphaFold 3 represents a transformative advancement in computational biology, enhancing protein structure prediction through novel multi-scale transformer architectures, biologically informed cross-attention mechanisms, and geometry-aware optimization strategies. These innovations dramatically improve predictive accuracy and generalization across diverse protein families, surpassing previous methods. Crucially, AlphaFold 3 embodies a paradigm shift toward differentiable simulation, bridging traditional static structural modeling with dynamic molecular simulations. By reframing protein folding predictions as a differentiable process, AlphaFold 3 serves as a foundational framework for integrating deep learning with physics-based molecular
Authors: Md. Rashid Shahriar Khan, Md. Abrar Hasan, Mohammod Tareq Aziz Justice
Abstract: Detecting anomalies in surveillance footage is inherently challenging due to their unpredictable and context-dependent nature. This work introduces a novel context-aware zero-shot anomaly detection framework that identifies abnormal events without exposure to anomaly examples during training. The proposed hybrid architecture combines TimeSformer, DPC, and CLIP to model spatiotemporal dynamics and semantic context. TimeSformer serves as the vision backbone to extract rich spatial-temporal features, while DPC forecasts future representations to identify temporal deviations. Furthermore, a CLIP-based semantic stream enables concept-level anomaly detection through context-specific text prompts. These components are jointly trained using InfoNCE and CPC losses, aligning visual inputs with their temporal and semantic representations. A context-gating mechanism further enhances decision-making by modulating predictions with scene-aware cues or global video features. By integrating predictive modeling with vision-language understanding, the system can generalize to previously unseen behaviors in complex environments. This framework bridges the gap between temporal reasoning and semantic context in zero-shot anomaly detection for surveillance. The code for this research has been made available at https://github.com/NK-II/Context-Aware-Zero-Shot-Anomaly-Detection-in-Surveillance.
URLs: https://github.com/NK-II/Context-Aware-Zero-Shot-Anomaly-Detection-in-Surveillance.
Authors: Ziqing Guo, Ziwen Pan, Alex Khan, Jan Balewski
Abstract: Vectorized quantum block encoding provides a way to embed classical data into Hilbert space, offering a pathway for quantum models, such as Quantum Transformers (QT), that replace classical self-attention with quantum circuit simulations to operate more efficiently. Current QTs rely on deep parameterized quantum circuits (PQCs), rendering them vulnerable to QPU noise, and thus hindering their practical performance. In this paper, we propose the Vectorized Quantum Transformer (VQT), a model that supports ideal masked attention matrix computation through quantum approximation simulation and efficient training via vectorized nonlinear quantum encoder, yielding shot-efficient and gradient-free quantum circuit simulation (QCS) and reduced classical sampling overhead. In addition, we demonstrate an accuracy comparison for IBM and IonQ in quantum circuit simulation and competitive results in benchmarking natural language processing tasks on IBM state-of-the-art and high-fidelity Kingston QPU. Our noise intermediate-scale quantum friendly VQT approach unlocks a novel architecture for end-to-end machine learning in quantum computing.
Authors: Jiawei Li, Akshayaa Magesh, Venugopal V. Veeravalli
Abstract: While Large Language Models (LLMs) have emerged as powerful foundational models to solve a variety of tasks, they have also been shown to be prone to hallucinations, i.e., generating responses that sound confident but are actually incorrect or even nonsensical. In this work, we formulate the problem of detecting hallucinations as a hypothesis testing problem and draw parallels to the problem of out-of-distribution detection in machine learning models. We propose a multiple-testing-inspired method to solve the hallucination detection problem, and provide extensive experimental results to validate the robustness of our approach against state-of-the-art methods.
Authors: Zekun Ni, Jonathan Weyn, Hang Zhang, Yanfei Xiang, Jiang Bian, Weixin Jin, Kit Thambiratnam, Qi Zhang, Haiyu Dong, Hongyu Sun
Abstract: Over the past few years, machine learning-based data-driven weather prediction has been transforming operational weather forecasting by providing more accurate forecasts while using a mere fraction of computing power compared to traditional numerical weather prediction (NWP). However, those models still rely on initial conditions from NWP, putting an upper limit on their forecast abilities. A few end-to-end systems have since been proposed, but they have yet to match the forecast skill of state-of-the-art NWP competitors. In this work, we propose Huracan, an observation-driven weather forecasting system which combines an ensemble data assimilation model with a forecast model to produce highly accurate forecasts relying only on observations as inputs. Huracan is not only the first to provide ensemble initial conditions and end-to-end ensemble weather forecasts, but also the first end-to-end system to achieve an accuracy comparable with that of ECMWF ENS, the state-of-the-art NWP competitor, despite using a smaller amount of available observation data. Notably, Huracan matches or exceeds the continuous ranked probability score of ECMWF ENS on 75.4% of the variable and lead time combinations. Our work is a major step forward in end-to-end data-driven weather prediction and opens up opportunities for further improving and revolutionizing operational weather forecasting.
Authors: Yusi Wei, Hande Y. Benson, Muge Capan
Abstract: The secondary use of healthcare data is vital for research and clinical innovation, but it raises concerns about patient privacy. This study investigates how to balance privacy preservation and data utility in healthcare data sharing, considering the perspectives of both data providers and data users. Using a dataset of adult patients hospitalized between 2013 and 2015, we predict whether sepsis was present at admission or developed during the hospital stay. We identify sub-populations, such as older adults, frequently hospitalized patients, and racial minorities, that are especially vulnerable to privacy attacks due to their unique combinations of demographic and healthcare utilization attributes. These groups are also critical for machine learning (ML) model performance. We evaluate three anonymization methods-$k$-anonymity, the technique by Zheng et al., and the MO-OBAM model-based on their ability to reduce re-identification risk while maintaining ML utility. Results show that $k$-anonymity offers limited protection. The methods of Zheng et al. and MO-OBAM provide stronger privacy safeguards, with MO-OBAM yielding the best utility outcomes: only a 2% change in precision and recall compared to the original dataset. This work provides actionable insights for healthcare organizations on how to share data responsibly. It highlights the need for anonymization methods that protect vulnerable populations without sacrificing the performance of data-driven models.
Authors: Jongyeong Lee, Junya Honda, Shinji Ito, Min-hwan Oh
Abstract: Follow-the-Regularized-Leader (FTRL) policies have achieved Best-of-Both-Worlds (BOBW) results in various settings through hybrid regularizers, whereas analogous results for Follow-the-Perturbed-Leader (FTPL) remain limited due to inherent analytical challenges. To advance the analytical foundations of FTPL, we revisit classical FTRL-FTPL duality for unbounded perturbations and establish BOBW results for FTPL under a broad family of asymmetric unbounded Fr\'echet-type perturbations, including hybrid perturbations combining Gumbel-type and Fr\'echet-type tails. These results not only extend the BOBW results of FTPL but also offer new insights into designing alternative FTPL policies competitive with hybrid regularization approaches. Motivated by earlier observations in two-armed bandits, we further investigate the connection between the $1/2$-Tsallis entropy and a Fr\'echet-type perturbation. Our numerical observations suggest that it corresponds to a symmetric Fr\'echet-type perturbation, and based on this, we establish the first BOBW guarantee for symmetric unbounded perturbations in the two-armed setting. In contrast, in general multi-armed bandits, we find an instance in which symmetric Fr\'echet-type perturbations violate the key condition for standard BOBW analysis, which is a problem not observed with asymmetric or nonnegative Fr\'echet-type perturbations. Although this example does not rule out alternative analyses achieving BOBW results, it suggests the limitations of directly applying the relationship observed in two-armed cases to the general case and thus emphasizes the need for further investigation to fully understand the behavior of FTPL in broader settings.
Authors: Chenxi Zhou, Pengfei Cao, Jiang Li, Jun Zhao, Kang Liu
Abstract: Large language models (LLMs) present significant deployment challenges due to their scale, with post-training quantization (PTQ) emerging as a practical compression solution. However, a comprehensive understanding of how PTQ precisely impacts diverse LLM knowledge capabilities remains elusive, and existing scaling laws for quantized models often overlook crucial PTQ-specific parameters and task-specific sensitivities. This paper addresses these gaps by conducting an extensive empirical investigation to establish task-stratified scaling laws. We disentangle LLM knowledge into memorization and utilization capabilities and develop a unified quantitative framework that incorporates model size, effective bit-width, calibration set size, and group size. Our central finding reveals that knowledge memorization exhibits markedly greater sensitivity to variations in effective bit-width, calibration set size, and model size compared to the more robust knowledge utilization. These findings offer a fine-grained understanding of PTQ's impact and provide guidance for developing knowledge-aware quantization strategies that can better preserve targeted cognitive functions.
Authors: Shrenik Jadhav, Birva Sevak, Srijita Das, Akhtar Hussain, Wencong Su, Van-Hai Bui
Abstract: Peer-to-peer (P2P) energy trading is becoming central to modern distribution systems as rooftop PV and home energy management systems become pervasive, yet most existing market and reinforcement learning designs emphasize efficiency or private profit and offer little real-time guidance to ensure equitable outcomes under uncertainty. To address this gap, a fairness-aware multiagent reinforcement learning framework, FairMarket-RL, is proposed in which a large language model (LLM) critic shapes bidding policies within a continuous double auction under partial observability and discrete price-quantity actions. After each trading slot, the LLM returns normalized fairness scores Fairness-to-Grid (FTG), Fairness-Between-Sellers (FBS), and Fairness-of-Pricing (FPP) that are integrated into the reward via ramped coefficients and tunable scaling, so that fairness guidance complements, rather than overwhelms, economic incentives. The environment models realistic residential load and PV profiles and enforce hard constraints on prices, physical feasibility, and policy-update stability. Across a progression of experiments from a small pilot to a larger simulated community and a mixed-asset real-world dataset, the framework shifts exchanges toward local P2P trades, lowers consumer costs relative to grid-only procurement, sustains strong fairness across participants, and preserves utility viability. Sensitivity analyses over solar availability and aggregate demand further indicate robust performance, suggesting a scalable, LLM-guided pathway to decentralized electricity markets that are economically efficient, socially equitable, and technically sound.
Authors: Soumen Ghosh, Christine Jestin Hannan, Rajat Vashistha, Parveen Kundu, Sandra Brosda, Lauren G. Aoude, James Lonie, Andrew Nathanson, Jessica Ng, Andrew P. Barbour, Viktor Vegh
Abstract: Robust generalization is essential for deploying deep learning based tumor segmentation in clinical PET-CT workflows, where anatomical sites, scanners, and patient populations vary widely. This study presents the first cross cancer evaluation of nnU-Net on PET-CT, introducing two novel, expert-annotated whole-body datasets. 279 patients with oesophageal cancer (Australian cohort) and 54 with lung cancer (Indian cohort). These cohorts complement the public AutoPET dataset and enable systematic stress-testing of cross domain performance. We trained and tested 3D nnUNet models under three paradigms. Target only (oesophageal), public only (AutoPET), and combined training. For the tested sets, the oesophageal only model achieved the best in-domain accuracy (mean DSC, 57.8) but failed on external Indian lung cohort (mean DSC less than 3.4), indicating severe overfitting. The public only model generalized more broadly (mean DSC, 63.5 on AutoPET, 51.6 on Indian lung cohort) but underperformed in oesophageal Australian cohort (mean DSC, 26.7). The combined approach provided the most balanced results (mean DSC, lung (52.9), oesophageal (40.7), AutoPET (60.9)), reducing boundary errors and improving robustness across all cohorts. These findings demonstrate that dataset diversity, particularly multi demographic, multi center and multi cancer integration, outweighs architectural novelty as the key driver of robust generalization. This work presents the demography based cross cancer deep learning segmentation evaluation and highlights dataset diversity, rather than model complexity, as the foundation for clinically robust segmentation.
Authors: Eichi Takaya, Ryusei Inamori
Abstract: Background and objective: Expert annotations limit large-scale supervised pretraining in medical imaging, while ubiquitous metadata (modality, anatomical region) remain underused. We introduce ModAn-MulSupCon, a modality- and anatomy-aware multi-label supervised contrastive pretraining method that leverages such metadata to learn transferable representations. Method: Each image's modality and anatomy are encoded as a multi-hot vector. A ResNet-18 encoder is pretrained on a mini subset of RadImageNet (miniRIN, 16,222 images) with a Jaccard-weighted multi-label supervised contrastive loss, and then evaluated by fine-tuning and linear probing on three binary classification tasks--ACL tear (knee MRI), lesion malignancy (breast ultrasound), and nodule malignancy (thyroid ultrasound). Result: With fine-tuning, ModAn-MulSupCon achieved the best AUC on MRNet-ACL (0.964) and Thyroid (0.763), surpassing all baselines ($p<0.05$), and ranked second on Breast (0.926) behind SimCLR (0.940; not significant). With the encoder frozen, SimCLR/ImageNet were superior, indicating that ModAn-MulSupCon representations benefit most from task adaptation rather than linear separability. Conclusion: Encoding readily available modality/anatomy metadata as multi-label targets provides a practical, scalable pretraining signal that improves downstream accuracy when fine-tuning is feasible. ModAn-MulSupCon is a strong initialization for label-scarce clinical settings, whereas SimCLR/ImageNet remain preferable for frozen-encoder deployments.
Authors: Chenxuan Miao, Yutong Feng, Jianshu Zeng, Zixiang Gao, Hantang Liu, Yunfeng Yan, Donglian Qi, Xi Chen, Bin Wang, Hengshuang Zhao
Abstract: Video object removal has achieved advanced performance due to the recent success of video generative models. However, when addressing the side effects of objects, e.g., their shadows and reflections, existing works struggle to eliminate these effects for the scarcity of paired video data as supervision. This paper presents ROSE, termed Remove Objects with Side Effects, a framework that systematically studies the object's effects on environment, which can be categorized into five common cases: shadows, reflections, light, translucency and mirror. Given the challenges of curating paired videos exhibiting the aforementioned effects, we leverage a 3D rendering engine for synthetic data generation. We carefully construct a fully-automatic pipeline for data preparation, which simulates a large-scale paired dataset with diverse scenes, objects, shooting angles, and camera trajectories. ROSE is implemented as an video inpainting model built on diffusion transformer. To localize all object-correlated areas, the entire video is fed into the model for reference-based erasing. Moreover, additional supervision is introduced to explicitly predict the areas affected by side effects, which can be revealed through the differential mask between the paired videos. To fully investigate the model performance on various side effect removal, we presents a new benchmark, dubbed ROSE-Bench, incorporating both common scenarios and the five special side effects for comprehensive evaluation. Experimental results demonstrate that ROSE achieves superior performance compared to existing video object erasing models and generalizes well to real-world video scenarios. The project page is https://rose2025-inpaint.github.io/.
Authors: Jiajie He, Yuechun Gu, Min-Chun Chen, Keke Chen
Abstract: Large language models (LLMs) based Recommender Systems (RecSys) can flexibly adapt recommendation systems to different domains. It utilizes in-context learning (ICL), i.e., the prompts, to customize the recommendation functions, which include sensitive historical user-specific item interactions, e.g., implicit feedback like clicked items or explicit product reviews. Such private information may be exposed to novel privacy attack. However, no study has been done on this important issue. We design four membership inference attacks (MIAs), aiming to reveal whether victims' historical interactions have been used by system prompts. They are \emph{direct inquiry, hallucination, similarity, and poisoning attacks}, each of which utilizes the unique features of LLMs or RecSys. We have carefully evaluated them on three LLMs that have been used to develop ICL-LLM RecSys and two well-known RecSys benchmark datasets. The results confirm that the MIA threat on LLM RecSys is realistic: direct inquiry and poisoning attacks showing significantly high attack advantages. We have also analyzed the factors affecting these attacks, such as the number of shots in system prompts and the position of the victim in the shots.
Authors: Shaswata Mitra, Azim Bazarov, Martin Duclos, Sudip Mittal, Aritran Piplai, Md Rayhanur Rahman, Edward Zieglar, Shahram Rahimi
Abstract: Signature-based Intrusion Detection Systems (IDS) detect malicious activities by matching network or host activity against predefined rules. These rules are derived from extensive Cyber Threat Intelligence (CTI), which includes attack signatures and behavioral patterns obtained through automated tools and manual threat analysis, such as sandboxing. The CTI is then transformed into actionable rules for the IDS engine, enabling real-time detection and prevention. However, the constant evolution of cyber threats necessitates frequent rule updates, which delay deployment time and weaken overall security readiness. Recent advancements in agentic systems powered by Large Language Models (LLMs) offer the potential for autonomous IDS rule generation with internal evaluation. We introduce FALCON, an autonomous agentic framework that generates deployable IDS rules from CTI data in real-time and evaluates them using built-in multi-phased validators. To demonstrate versatility, we target both network (Snort) and host-based (YARA) mediums and construct a comprehensive dataset of IDS rules with their corresponding CTIs. Our evaluations indicate FALCON excels in automatic rule generation, with an average of 95% accuracy validated by qualitative evaluation with 84% inter-rater agreement among multiple cybersecurity analysts across all metrics. These results underscore the feasibility and effectiveness of LLM-driven data mining for real-time cyber threat mitigation.
Authors: Yi-Ping Hsu, Po-Wei Wang, Chantat Eksombatchai, Jiajing Xu
Abstract: ID-based embeddings are widely used in web-scale online recommendation systems. However, their susceptibility to overfitting, particularly due to the long-tail nature of data distributions, often limits training to a single epoch, a phenomenon known as the "one-epoch problem." This challenge has driven research efforts to optimize performance within the first epoch by enhancing convergence speed or feature sparsity. In this study, we introduce a novel two-stage training strategy that incorporates a pre-training phase using a minimal model with contrastive loss, enabling broader data coverage for the embedding system. Our offline experiments demonstrate that multi-epoch training during the pre-training phase does not lead to overfitting, and the resulting embeddings improve online generalization when fine-tuned for more complex downstream recommendation tasks. We deployed the proposed system in live traffic at Pinterest, achieving significant site-wide engagement gains.
Authors: Yu Kamimizu, Hao Liu, Toshiyuki Nakata
Abstract: Insects control unsteady aerodynamic forces on flapping wings to navigate complex environments. While understanding these forces is vital for biology, physics, and engineering, existing evaluation methods face trade-offs: high-fidelity simulations are computationally or experimentally expensive and lack explanatory power, whereas theoretical models based on quasi-steady assumptions offer insights but exhibit low accuracy. To overcome these limitations and thus enhance the accuracy of quasi-steady aerodynamic models, we applied a data-driven approach involving discovery and formulation of previously overlooked critical mechanisms. Through selection from 5,000 candidate kinematic functions, we identified mathematical expressions for three key additional mechanisms -- the effect of advance ratio, effect of spanwise kinematic velocity, and rotational Wagner effect -- which had been qualitatively recognized but were not formulated. Incorporating these mechanisms considerably reduced the prediction errors of the quasi-steady model using the computational fluid dynamics results as the ground truth, both in hawkmoth forward flight (at high Reynolds numbers) and fruit fly maneuvers (at low Reynolds numbers). The data-driven quasi-steady model enables rapid aerodynamic analysis, serving as a practical tool for understanding evolutionary adaptations in insect flight and developing bio-inspired flying robots.
Authors: Promise Osaine Ekpo, Brian La, Thomas Wiener, Saesha Agarwal, Arshia Agrawal, Gonzalo Gonzalez-Pumariega, Lekan P. Molu, Angelique Taylor
Abstract: Fairness in multi-agent reinforcement learning (MARL) is often framed as a workload balance problem, overlooking agent expertise and the structured coordination required in real-world domains. In healthcare, equitable task allocation requires workload balance or expertise alignment to prevent burnout and overuse of highly skilled agents. Workload balance refers to distributing an approximately equal number of subtasks or equalised effort across healthcare workers, regardless of their expertise. We make two contributions to address this problem. First, we propose FairSkillMARL, a framework that defines fairness as the dual objective of workload balance and skill-task alignment. Second, we introduce MARLHospital, a customizable healthcare-inspired environment for modeling team compositions and energy-constrained scheduling impacts on fairness, as no existing simulators are well-suited for this problem. We conducted experiments to compare FairSkillMARL in conjunction with four standard MARL methods, and against two state-of-the-art fairness metrics. Our results suggest that fairness based solely on equal workload might lead to task-skill mismatches and highlight the need for more robust metrics that capture skill-task misalignment. Our work provides tools and a foundation for studying fairness in heterogeneous multi-agent systems where aligning effort with expertise is critical.
Authors: Melanie Wille, Tobias Fischer, Scarlett Raine
Abstract: Underwater object detection is critical for monitoring marine ecosystems but poses unique challenges, including degraded image quality, imbalanced class distribution, and distinct visual characteristics. Not every species is detected equally well, yet underlying causes remain unclear. We address two key research questions: 1) What factors beyond data quantity drive class-specific performance disparities? 2) How can we systematically improve detection of under-performing marine species? We manipulate the DUO dataset to separate the object detection task into localization and classification and investigate the under-performance of the scallop class. Localization analysis using YOLO11 and TIDE finds that foreground-background discrimination is the most problematic stage regardless of data quantity. Classification experiments reveal persistent precision gaps even with balanced data, indicating intrinsic feature-based challenges beyond data scarcity and inter-class dependencies. We recommend imbalanced distributions when prioritizing precision, and balanced distributions when prioritizing recall. Improving under-performing classes should focus on algorithmic advances, especially within localization modules. We publicly release our code and datasets.
Authors: Jungwoo Kim, Minsang Kim, Jaeheon Lee, Chanwoo Moon, Heejin Kim, Taeho Hwang, Woosuk Chung, Yeseong Kim, Sungjin Lee
Abstract: Serving Large Language Models (LLMs) at scale requires meeting strict Service Level Objectives (SLOs) under severe computational and memory constraints. Nevertheless, traditional caching strategies fall short: exact-matching and prefix caches neglect query semantics, while state-of-the-art semantic caches remain confined to traditional intuitions, offering little conceptual departure. Building on this, we present SISO, a semantic caching system that redefines efficiency for LLM serving. SISO introduces centroid-based caching to maximize coverage with minimal memory, locality-aware replacement to preserve high-value entries, and dynamic thresholding to balance accuracy and latency under varying workloads. Across diverse datasets, SISO delivers up to 1.71$\times$ higher hit ratios and consistently stronger SLO attainment compared to state-of-the-art systems.
Authors: Chang Wang, Siyu Yan, Depeng Yuan, Yuqi Chen, Yanhua Huang, Yuanhang Zheng, Shuhao Li, Yinqi Zhang, Kedi Chen, Mingrui Zhu, Ruiwen Xu
Abstract: The generation of ad headlines plays a vital role in modern advertising, where both quality and diversity are essential to engage a broad range of audience segments. Current approaches primarily optimize language models for headline quality or click-through rates (CTR), often overlooking the need for diversity and resulting in homogeneous outputs. To address this limitation, we propose DIVER, a novel framework based on large language models (LLMs) that are jointly optimized for both diversity and quality. We first design a semantic- and stylistic-aware data generation pipeline that automatically produces high-quality training pairs with ad content and multiple diverse headlines. To achieve the goal of generating high-quality and diversified ad headlines within a single forward pass, we propose a multi-stage multi-objective optimization framework with supervised fine-tuning (SFT) and reinforcement learning (RL). Experiments on real-world industrial datasets demonstrate that DIVER effectively balances quality and diversity. Deployed on a large-scale content-sharing platform serving hundreds of millions of users, our framework improves advertiser value (ADVV) and CTR by 4.0% and 1.4%.
Authors: Mengmeng Li, Philipp Schneider, Jelisaveta Aleksi\'c, Daniel Kuhn
Abstract: We introduce the first best-of-both-worlds algorithm for contextual combinatorial semi-bandits that simultaneously guarantees $\widetilde{\mathcal{O}}(\sqrt{T})$ regret in the adversarial regime and $\widetilde{\mathcal{O}}(\ln T)$ regret in the corrupted stochastic regime. Our approach builds on the Follow-the-Regularized-Leader (FTRL) framework equipped with a Shannon entropy regularizer, yielding a flexible method that admits efficient implementations. Beyond regret bounds, we tackle the practical bottleneck in FTRL (or, equivalently, Online Stochastic Mirror Descent) arising from the high-dimensional projection step encountered in each round of interaction. By leveraging the Karush-Kuhn-Tucker conditions, we transform the $K$-dimensional convex projection problem into a single-variable root-finding problem, dramatically accelerating each round. Empirical evaluations demonstrate that this combined strategy not only attains the attractive regret bounds of best-of-both-worlds algorithms but also delivers substantial per-round speed-ups, making it well-suited for large-scale, real-time applications.
Authors: Christian L\"owens, Thorben Funke, Jingchao Xie, Alexandru Paul Condurache
Abstract: Online mapping models show remarkable results in predicting vectorized maps from multi-view camera images only. However, all existing approaches still rely on ground-truth high-definition maps during training, which are expensive to obtain and often not geographically diverse enough for reliable generalization. In this work, we propose PseudoMapTrainer, a novel approach to online mapping that uses pseudo-labels generated from unlabeled sensor data. We derive those pseudo-labels by reconstructing the road surface from multi-camera imagery using Gaussian splatting and semantics of a pre-trained 2D segmentation network. In addition, we introduce a mask-aware assignment algorithm and loss function to handle partially masked pseudo-labels, allowing for the first time the training of online mapping models without any ground-truth maps. Furthermore, our pseudo-labels can be effectively used to pre-train an online model in a semi-supervised manner to leverage large-scale unlabeled crowdsourced data. The code is available at github.com/boschresearch/PseudoMapTrainer.
Authors: Yannick Hollenweger, Dennis M. Kochman, Burigede Liu
Abstract: Neural network surrogate models for constitutive laws in computational mechanics have been in use for some time. In plasticity, these models often rely on gated recurrent units (GRUs) or long short-term memory (LSTM) cells, which excel at capturing path-dependent phenomena. However, they suffer from long training times and time-resolution-dependent predictions that extrapolate poorly. Moreover, most existing surrogates for macro- or mesoscopic plasticity handle only relatively simple material behavior. To overcome these limitations, we introduce the Temperature-Aware Recurrent Neural Operator (TRNO), a time-resolution-independent neural architecture. We apply the TRNO to model the temperature-dependent plastic response of polycrystalline magnesium, which shows strong plastic anisotropy and thermal sensitivity. The TRNO achieves high predictive accuracy and generalizes effectively across diverse loading cases, temperatures, and time resolutions. It also outperforms conventional GRU and LSTM models in training efficiency and predictive performance. Finally, we demonstrate multiscale simulations with the TRNO, yielding a speedup of at least three orders of magnitude over traditional constitutive models.
Authors: Colin Merk, Ismail Geles, Jiaxu Xing, Angel Romero, Giorgia Ramponi, Davide Scaramuzza
Abstract: Preference-based reinforcement learning (PbRL) enables agents to learn control policies without requiring manually designed reward functions, making it well-suited for tasks where objectives are difficult to formalize or inherently subjective. Acrobatic flight poses a particularly challenging problem due to its complex dynamics, rapid movements, and the importance of precise execution. In this work, we explore the use of PbRL for agile drone control, focusing on the execution of dynamic maneuvers such as powerloops. Building on Preference-based Proximal Policy Optimization (Preference PPO), we propose Reward Ensemble under Confidence (REC), an extension to the reward learning objective that improves preference modeling and learning stability. Our method achieves 88.4% of the shaped reward performance, compared to 55.2% with standard Preference PPO. We train policies in simulation and successfully transfer them to real-world drones, demonstrating multiple acrobatic maneuvers where human preferences emphasize stylistic qualities of motion. Furthermore, we demonstrate the applicability of our probabilistic reward model in a representative MuJoCo environment for continuous control. Finally, we highlight the limitations of manually designed rewards, observing only 60.7% agreement with human preferences. These results underscore the effectiveness of PbRL in capturing complex, human-centered objectives across both physical and simulated domains.
Authors: Viktor N. Zhuravlev, Artur R. Khairullin, Ernest A. Dyagin, Alena N. Sitkina, Nikita I. Kulin
Abstract: Autoprompting is the process of automatically selecting optimized prompts for language models, which has been gaining popularity with the rapid advancement of prompt engineering, driven by extensive research in the field of large language models (LLMs). This paper presents ReflectivePrompt - a novel autoprompting method based on evolutionary algorithms that employs a reflective evolution approach for more precise and comprehensive search of optimal prompts. ReflectivePrompt utilizes short-term and long-term reflection operations before crossover and elitist mutation to enhance the quality of the modifications they introduce. This method allows for the accumulation of knowledge obtained throughout the evolution process and updates it at each epoch based on the current population. ReflectivePrompt was tested on 33 datasets for classification and text generation tasks using open-access large language models: t-lite-instruct-0.1 and gemma3-27b-it. The method demonstrates, on average, a significant improvement (e.g., 28% on BBH compared to EvoPrompt) in metrics relative to current state-of-the-art approaches, thereby establishing itself as one of the most effective solutions in evolutionary algorithm-based autoprompting.
Authors: Kushagra Agrawal, Nisharg Nargund, Oishani Banerjee
Abstract: Vector similarity search plays a pivotal role in modern information retrieval systems, especially when powered by transformer-based embeddings. However, the scalability and efficiency of such systems are often hindered by the high dimensionality of latent representations. In this paper, we propose a novel game-theoretic framework for optimizing latent-space compression to enhance both the efficiency and semantic utility of vector search. By modeling the compression strategy as a zero-sum game between retrieval accuracy and storage efficiency, we derive a latent transformation that preserves semantic similarity while reducing redundancy. We benchmark our method against FAISS, a widely-used vector search library, and demonstrate that our approach achieves a significantly higher average similarity (0.9981 vs. 0.5517) and utility (0.8873 vs. 0.5194), albeit with a modest increase in query time. This trade-off highlights the practical value of game-theoretic latent compression in high-utility, transformer-based search applications. The proposed system can be seamlessly integrated into existing LLM pipelines to yield more semantically accurate and computationally efficient retrieval.
Authors: Mona Mirzaie, Bodo Rosenhahn
Abstract: Trustworthy AI is mandatory for the broad deployment of autonomous vehicles. Although end-to-end approaches derive control commands directly from raw data, interpreting these decisions remains challenging, especially in complex urban scenarios. This is mainly attributed to very deep neural networks with non-linear decision boundaries, making it challenging to grasp the logic behind AI-driven decisions. This paper presents a method to enhance interpretability while optimizing control commands in autonomous driving. To address this, we propose loss functions that promote the interpretability of our model by generating sparse and localized feature maps. The feature activations allow us to explain which image regions contribute to the predicted control command. We conduct comprehensive ablation studies on the feature extraction step and validate our method on the CARLA benchmarks. We also demonstrate that our approach improves interpretability, which correlates with reducing infractions, yielding a safer, high-performance driving model. Notably, our monocular, non-ensemble model surpasses the top-performing approaches from the CARLA Leaderboard by achieving lower infraction scores and the highest route completion rate, all while ensuring interpretability.
Authors: Peter Naylor, Benjamin Poignard, H\'ector Climente-Gonz\'alez, Makoto Yamada
Abstract: We propose a feature screening method that integrates both feature-feature and feature-target relationships. Inactive features are identified via a penalized minimum Redundancy Maximum Relevance (mRMR) procedure, which is the continuous version of the classic mRMR penalized by a non-convex regularizer, and where the parameters estimated as zero coefficients represent the set of inactive features. We establish the conditions under which zero coefficients are correctly identified to guarantee accurate recovery of inactive features. We introduce a multi-stage procedure based on the knockoff filter enabling the penalized mRMR to discard inactive features while controlling the false discovery rate (FDR). Our method performs comparably to HSIC-LASSO but is more conservative in the number of selected features. It only requires setting an FDR threshold, rather than specifying the number of features to retain. The effectiveness of the method is illustrated through simulations and real-world datasets. The code to reproduce this work is available on the following GitHub: https://github.com/PeterJackNaylor/SmRMR.
Authors: Mahmoud Dhimish
Abstract: Thermal anomaly detection in solar photovoltaic (PV) systems is essential for ensuring operational efficiency and reducing maintenance costs. In this study, we developed and named HOTSPOT-YOLO, a lightweight artificial intelligence (AI) model that integrates an efficient convolutional neural network backbone and attention mechanisms to improve object detection. This model is specifically designed for drone-based thermal inspections of PV systems, addressing the unique challenges of detecting small and subtle thermal anomalies, such as hotspots and defective modules, while maintaining real-time performance. Experimental results demonstrate a mean average precision of 90.8%, reflecting a significant improvement over baseline object detection models. With a reduced computational load and robustness under diverse environmental conditions, HOTSPOT-YOLO offers a scalable and reliable solution for large-scale PV inspections. This work highlights the integration of advanced AI techniques with practical engineering applications, revolutionizing automated fault detection in renewable energy systems.
Authors: Jakub Micha\'nk\'ow
Abstract: This study evaluates deep neural networks for forecasting probability distributions of financial returns. 1D convolutional neural networks (CNN) and Long Short-Term Memory (LSTM) architectures are used to forecast parameters of three probability distributions: Normal, Student's t, and skewed Student's t. Using custom negative log-likelihood loss functions, distribution parameters are optimized directly. The models are tested on six major equity indices (S\&P 500, BOVESPA, DAX, WIG, Nikkei 225, and KOSPI) using probabilistic evaluation metrics including Log Predictive Score (LPS), Continuous Ranked Probability Score (CRPS), and Probability Integral Transform (PIT). Results show that deep learning models provide accurate distributional forecasts and perform competitively with classical GARCH models for Value-at-Risk estimation. The LSTM with skewed Student's t distribution performs best across multiple evaluation criteria, capturing both heavy tails and asymmetry in financial returns. This work shows that deep neural networks are viable alternatives to traditional econometric models for financial risk assessment and portfolio management.
Authors: Rodrigo Carmo Terin
Abstract: The development of a Euclidean stochastic field-theoretic approach that maps deep neural networks (DNNs) to quantum electrodynamics (QED) with local U(1) symmetry is presented. Neural activations and weights are represented by fermionic matter and gauge fields, with a fictitious Langevin time enabling covariant gauge fixing. This mapping identifies the gauge parameter with kernel design choices in wide DNNs, relating stability thresholds to gauge-dependent amplification factors. Finite-width fluctuations correspond to loop corrections in QED. As a proof of concept, we validate the theoretical predictions through numerical simulations of standard multilayer perceptrons and, in parallel, propose a gauge-invariant neural network (GINN) implementation using magnitude--phase parameterization of weights. Finally, a double-copy replica approach is shown to unify the computation of the largest Lyapunov exponent in stochastic QED and wide DNNs.
Authors: Simpenzwe Honore Leandre, Natenaile Asmamaw Shiferaw, Dillip Rout
Abstract: In this paper, we propose a vision model that adopts token mixing, sequence-pooling, and convolutional tokenizers to achieve state-of-the-art performance and efficient inference in fixed context-length tasks. In the CIFAR100 benchmark, our model significantly improves the baseline of the top 1% and top 5% validation accuracy from 36.50% to 46.29% and 66.33% to 76.31%, while being more efficient than the Scaled Dot Product Attention (SDPA) transformers when the context length is less than the embedding dimension and only 60% the size. In addition, the architecture demonstrates high training stability and does not rely on techniques such as data augmentation like mixup, positional embeddings, or learning rate scheduling. We make our code available on Github.
Authors: Shaojin Wu, Mengqi Huang, Yufeng Cheng, Wenxu Wu, Jiahe Tian, Yiming Luo, Fei Ding, Qian He
Abstract: Existing literature typically treats style-driven and subject-driven generation as two disjoint tasks: the former prioritizes stylistic similarity, whereas the latter insists on subject consistency, resulting in an apparent antagonism. We argue that both objectives can be unified under a single framework because they ultimately concern the disentanglement and re-composition of content and style, a long-standing theme in style-driven research. To this end, we present USO, a Unified Style-Subject Optimized customization model. First, we construct a large-scale triplet dataset consisting of content images, style images, and their corresponding stylized content images. Second, we introduce a disentangled learning scheme that simultaneously aligns style features and disentangles content from style through two complementary objectives, style-alignment training and content-style disentanglement training. Third, we incorporate a style reward-learning paradigm denoted as SRL to further enhance the model's performance. Finally, we release USO-Bench, the first benchmark that jointly evaluates style similarity and subject fidelity across multiple metrics. Extensive experiments demonstrate that USO achieves state-of-the-art performance among open-source models along both dimensions of subject consistency and style similarity. Code and model: https://github.com/bytedance/USO
Authors: Hung Ming Liu
Abstract: We present a framework where neural models develop an AI Mother Tongue, a native symbolic language that simultaneously supports intuitive reasoning, compositional symbol chains, and inherent interpretability. Unlike post-hoc explanation methods, our approach embeds reasoning directly into the model's representations: symbols capture meaningful semantic patterns, chains trace decision paths, and gated induction mechanisms guide selective focus, yielding transparent yet flexible reasoning. We introduce complementary training objectives to enhance symbol purity and decision sparsity, and employ a sequential specialization strategy to first build broad symbolic competence and then refine intuitive judgments. Experiments on AI tasks demonstrate competitive accuracy alongside verifiable reasoning traces, showing that AI Mother Tongue can serve as a unified mechanism for interpretability, intuition, and symbolic reasoning in neural models.
Authors: Viktor N. Zhuravlev, Artur R. Khairullin, Ernest A. Dyagin, Alena N. Sitkina, Nikita I. Kulin
Abstract: Autoprompting is the process of automatically selecting optimized prompts for language models, which is gaining popularity due to the rapid development of prompt engineering driven by extensive research in the field of large language models (LLMs). This paper presents DistillPrompt -- a novel autoprompting method based on large language models that employs a multi-stage integration of task-specific information into prompts using training data. DistillPrompt utilizes distillation, compression, and aggregation operations to explore the prompt space more thoroughly. The method was tested on different datasets for text classification and generation tasks using the t-lite-instruct-0.1 language model. The results demonstrate a significant average improvement (e.g., 20.12% across the entire dataset compared to Grips) in key metrics over existing methods in the field, establishing DistillPrompt as one of the most effective non-gradient approaches in autoprompting.
Authors: Shanyan Lai
Abstract: This study investigates the pretrained RNN attention models with the mainstream attention mechanisms such as additive attention, Luong's three attentions, global self-attention (Self-att) and sliding window sparse attention (Sparse-att) for the empirical asset pricing research on top 420 large-cap US stocks. This is the first paper on the large-scale state-of-the-art (SOTA) attention mechanisms applied in the asset pricing context. They overcome the limitations of the traditional machine learning (ML) based asset pricing, such as mis-capturing the temporal dependency and short memory. Moreover, the enforced causal masks in the attention mechanisms address the future data leaking issue ignored by the more advanced attention-based models, such as the classic Transformer. The proposed attention models also consider the temporal sparsity characteristic of asset pricing data and mitigate potential overfitting issues by deploying the simplified model structures. This provides some insights for future empirical economic research. All models are examined in three periods, which cover pre-COVID-19 (mild uptrend), COVID-19 (steep uptrend with a large drawdown) and one year post-COVID-19 (sideways movement with high fluctuations), for testing the stability of these models under extreme market conditions. The study finds that in value-weighted portfolio back testing, Model Self-att and Model Sparse-att exhibit great capabilities in deriving the absolute returns and hedging downside risks, while they achieve an annualized Sortino ratio of 2.0 and 1.80 respectively in the period with COVID-19. And Model Sparse-att performs more stably than Model Self-att from the perspective of absolute portfolio returns with respect to the size of stocks' market capitalization.
Authors: Julian Suk, Jolanda J. Wentzel, Patryk Rygiel, Joost Daemen, Daniel Rueckert, Jelmer M. Wolterink
Abstract: Leveraging big data for patient care is promising in many medical fields such as cardiovascular health. For example, hemodynamic biomarkers like wall shear stress could be assessed from patient-specific medical images via machine learning algorithms, bypassing the need for time-intensive computational fluid simulation. However, it is extremely challenging to amass large-enough datasets to effectively train such models. We could address this data scarcity by means of self-supervised pre-training and foundations models given large datasets of geometric artery models. In the context of coronary arteries, leveraging learned representations to improve hemodynamic biomarker assessment has not yet been well studied. In this work, we address this gap by investigating whether a large dataset (8449 shapes) consisting of geometric models of 3D blood vessels can benefit wall shear stress assessment in coronary artery models from a small-scale clinical trial (49 patients). We create a self-supervised target for the 3D blood vessels by computing the heat kernel signature, a quantity obtained via Laplacian eigenvectors, which captures the very essence of the shapes. We show how geometric representations learned from this datasets can boost segmentation of coronary arteries into regions of low, mid and high (time-averaged) wall shear stress even when trained on limited data.
Authors: Serban C. Tudosie, Alexander Denker, Zeljko Kereta, Simon Arridge
Abstract: Single-Pixel Imaging enables reconstructing objects using a single detector through sequential illuminations with structured light patterns. We propose a bilevel optimisation method for learning task-specific, binary illumination patterns, optimised for applications like single-pixel fluorescence microscopy. We address the non-differentiable nature of binary pattern optimisation using the Straight-Through Estimator and leveraging a Total Deep Variation regulariser in the bilevel formulation. We demonstrate our method on the CytoImageNet microscopy dataset and show that learned patterns achieve superior reconstruction performance compared to baseline methods, especially in highly undersampled regimes.
Authors: Sidahmed Benabderrahmane, Talal Rahwan
Abstract: Advanced Persistent Threats (APTs) represent a growing menace to modern digital infrastructure. Unlike traditional cyberattacks, APTs are stealthy, adaptive, and long-lasting, often bypassing signature-based detection systems. This paper introduces a novel framework for APT detection that unites deep learning, reinforcement learning (RL), and active learning into a cohesive, adaptive defense system. Our system combines auto-encoders for latent behavioral encoding with a multi-agent ensemble of RL-based defenders, each trained to distinguish between benign and malicious process behaviors. We identify a critical challenge in existing detection systems: their static nature and inability to adapt to evolving attack strategies. To this end, our architecture includes multiple RL agents (Q-Learning, PPO, DQN, adversarial defenders), each analyzing latent vectors generated by an auto-encoder. When any agent is uncertain about its decision, the system triggers an active learning loop to simulate expert feedback, thus refining decision boundaries. An ensemble voting mechanism, weighted by each agent's performance, ensures robust final predictions.
Authors: Ehsan Yousefzadeh-Asl-Miandoab, Reza Karimzadeh, Bulat Ibragimov, Florina M. Ciorba, P{\i}nar T\"oz\"un
Abstract: Studies conducted on enterprise-scale infrastructure have shown that GPUs -- the core computational resource for deep learning (DL) training -- are often significantly underutilized. DL task collocation on GPUs is an opportunity to address this challenge. However, it may result in (1) out-of-memory crashes for the subsequently arriving task and (2) slowdowns for all tasks sharing the GPU due to resource interference. The former challenge poses a threat to robustness, while the latter affects the quality of service and energy efficiency. We propose CARMA, a server-scale task-level collocation-aware resource management system that handles both collocation challenges. CARMA encompasses GPUMemNet, a novel ML-based GPU memory estimator framework for DL training tasks, to minimize out-of-memory errors and introduces collocation policies that cap GPU utilization to minimize interference. Furthermore, CARMA introduces a recovery method to ensure robust restart of tasks that crash. Our evaluation on traces modeled after real-world DL training task traces shows that CARMA increases the GPU utilization over time by 39.3\%, decreases the end-to-end execution time by $\sim$26.7\%, and reduces the GPU energy use by $\sim$14.2\%.
Authors: Hong-Ye Hu, Abigail McClain Gomez, Liyuan Chen, Aaron Trowbridge, Andy J. Goldschmidt, Zachary Manchester, Frederic T. Chong, Arthur Jaffe, Susanne F. Yelin
Abstract: Analog quantum simulators with global control fields have emerged as powerful platforms for exploring complex quantum phenomena. Recent breakthroughs, such as the coherent control of thousands of atoms, highlight the growing potential for quantum applications at scale. Despite these advances, a fundamental theoretical question remains unresolved: to what extent can such systems realize universal quantum dynamics under global control? Here we establish a necessary and sufficient condition for universal quantum computation using only global pulse control, proving that a broad class of analog quantum simulators is, in fact, universal. We further extend this framework to fermionic and bosonic systems, including modern platforms such as ultracold atoms in optical superlattices. Crucially, to connect the theoretical possibility with experimental reality, we introduce a new control technique into the experiment - direct quantum optimal control. This method enables the synthesis of complex effective Hamiltonians and allows us to incorporate realistic hardware constraints. To show its practical power, we experimentally engineer three-body interactions outside the blockade regime and demonstrate topological dynamics on a Rydberg atom array. Using the new control framework, we overcome key experimental challenges, including hardware limitations and atom position fluctuations in the non-blockade regime, by identifying smooth, short-duration pulses that achieve high-fidelity dynamics. Experimental measurements reveal dynamical signatures of symmetry-protected-topological edge modes, confirming both the expressivity and feasibility of our approach. Our work opens a new avenue for quantum simulation beyond native hardware Hamiltonians, enabling the engineering of effective multi-body interactions and advancing the frontier of quantum information processing with globally-controlled analog platforms.
Authors: Aneesh Rangnekar, Harini Veeraraghavan
Abstract: Accurate detection and segmentation of cancerous lesions from computed tomography (CT) scans is essential for automated treatment planning and cancer treatment response assessment. Transformer-based models with self-supervised pretraining can produce reliably accurate segmentation from in-distribution (ID) data but degrade when applied to out-of-distribution (OOD) datasets. We address this challenge with RF-Deep, a random forest classifier that utilizes deep features from a pretrained transformer encoder of the segmentation model to detect OOD scans and enhance segmentation reliability. The segmentation model comprises a Swin Transformer encoder, pretrained with masked image modeling (SimMIM) on 10,432 unlabeled 3D CT scans covering cancerous and non-cancerous conditions, with a convolution decoder, trained to segment lung cancers in 317 3D scans. Independent testing was performed on 603 3D CT public datasets that included one ID dataset and four OOD datasets comprising chest CTs with pulmonary embolism (PE) and COVID-19, and abdominal CTs with kidney cancers and healthy volunteers. RF-Deep detected OOD cases with a FPR95 of 18.26%, 27.66%, and less than 0.1% on PE, COVID-19, and abdominal CTs, consistently outperforming established OOD approaches. The RF-Deep classifier provides a simple and effective approach to enhance reliability of cancer segmentation in ID and OOD scenarios.
Authors: Florian Hahlbohm, Linus Franke, Leon Overk\"amping, Paula Wespe, Susana Castillo, Martin Eisemann, Marcus Magnor
Abstract: Implicit Neural Point Cloud (INPC) is a recent hybrid representation that combines the expressiveness of neural fields with the efficiency of point-based rendering, achieving state-of-the-art image quality in novel view synthesis. However, as with other high-quality approaches that query neural networks during rendering, the practical usability of INPC is limited by comparatively slow rendering. In this work, we present a collection of optimizations that significantly improve both the training and inference performance of INPC without sacrificing visual fidelity. The most significant modifications are an improved rasterizer implementation, more effective sampling techniques, and the incorporation of pre-training for the convolutional neural network used for hole-filling. Furthermore, we demonstrate that points can be modeled as small Gaussians during inference to further improve quality in extrapolated, e.g., close-up views of the scene. We design our implementations to be broadly applicable beyond INPC and systematically evaluate each modification in a series of experiments. Our optimized INPC pipeline achieves up to 25% faster training, 2x faster rendering, and 20% reduced VRAM usage paired with slight image quality improvements.
Authors: Juan-Pablo Ortega, Florian Rossmannek
Abstract: Recurrent neural networks (RNNs) have become increasingly popular in information processing tasks involving time series and temporal data. A fundamental property of RNNs is their ability to create reliable input/output responses, often linked to how the network handles its memory of the information it processed. Various notions have been proposed to conceptualize the behavior of memory in RNNs, including steady states, echo states, state forgetting, input forgetting, and fading memory. Although these notions are often used interchangeably, their precise relationships remain unclear. This work aims to unify these notions in a common language, derive new implications and equivalences between them, and provide alternative proofs to some existing results. By clarifying the relationships between these concepts, this research contributes to a deeper understanding of RNNs and their temporal information processing capabilities.
Authors: Chiu-Chou Lin
Abstract: Contemporary artificial intelligence (AI) development largely centers on rational decision-making, valued for its measurability and suitability for objective evaluation. Yet in real-world contexts, an intelligent agent's decisions are shaped not only by logic but also by deeper influences such as beliefs, values, and preferences. The diversity of human decision-making styles emerges from these differences, highlighting that "style" is an essential but often overlooked dimension of intelligence. This dissertation introduces playstyle as an alternative lens for observing and analyzing the decision-making behavior of intelligent agents, and examines its foundational meaning and historical context from a philosophical perspective. By analyzing how beliefs and values drive intentions and actions, we construct a two-tier framework for style formation: the external interaction loop with the environment and the internal cognitive loop of deliberation. On this basis, we formalize style-related characteristics and propose measurable indicators such as style capacity, style popularity, and evolutionary dynamics. The study focuses on three core research directions: (1) Defining and measuring playstyle, proposing a general playstyle metric based on discretized state spaces, and extending it to quantify strategic diversity and competitive balance; (2) Expressing and generating playstyle, exploring how reinforcement learning and imitation learning can be used to train agents exhibiting specific stylistic tendencies, and introducing a novel approach for human-like style learning and modeling; and (3) Practical applications, analyzing the potential of these techniques in domains such as game design and interactive entertainment. Finally, the dissertation outlines future extensions, including the role of style as a core element in building artificial general intelligence (AGI).
Authors: Rafael Sterzinger, Tingyu Lin, Robert Sablatnig
Abstract: A foundational task for the digital analysis of documents is text line segmentation. However, automating this process with deep learning models is challenging because it requires large, annotated datasets that are often unavailable for historical documents. Additionally, the annotation process is a labor- and cost-intensive task that requires expert knowledge, which makes few-shot learning a promising direction for reducing data requirements. In this work, we demonstrate that small and simple architectures, coupled with a topology-aware loss function, are more accurate and data-efficient than more complex alternatives. We pair a lightweight UNet++ with a connectivity-aware loss, initially developed for neuron morphology, which explicitly penalizes structural errors like line fragmentation and unintended line merges. To increase our limited data, we train on small patches extracted from a mere three annotated pages per manuscript. Our methodology significantly improves upon the current state-of-the-art on the U-DIADS-TL dataset, with a 200% increase in Recognition Accuracy and a 75% increase in Line Intersection over Union. Our method also achieves an F-Measure score on par with or even exceeding that of the competition winner of the DIVA-HisDB baseline detection task, all while requiring only three annotated pages, exemplifying the efficacy of our approach. Our implementation is publicly available at: https://github.com/RafaelSterzinger/acpr_few_shot_hist.
URLs: https://github.com/RafaelSterzinger/acpr_few_shot_hist.
Authors: Luca Grillotti (AIRL, Imperial College London), Lisa Coiffard (AIRL, Imperial College London), Oscar Pang (AIRL, Imperial College London), Maxence Faldor (AIRL, Imperial College London), Antoine Cully (AIRL, Imperial College London)
Abstract: Autonomous skill discovery aims to enable robots to acquire diverse behaviors without explicit supervision. Learning such behaviors directly on physical hardware remains challenging due to safety and data efficiency constraints. Existing methods, including Quality-Diversity Actor-Critic (QDAC), require manually defined skill spaces and carefully tuned heuristics, limiting real-world applicability. We propose Unsupervised Real-world Skill Acquisition (URSA), an extension of QDAC that enables robots to autonomously discover and master diverse, high-performing skills directly in the real world. We demonstrate that URSA successfully discovers diverse locomotion skills on a Unitree A1 quadruped in both simulation and the real world. Our approach supports both heuristic-driven skill discovery and fully unsupervised settings. We also show that the learned skill repertoire can be reused for downstream tasks such as real-world damage adaptation, where URSA outperforms all baselines in 5 out of 9 simulated and 3 out of 5 real-world damage scenarios. Our results establish a new framework for real-world robot learning that enables continuous skill discovery with limited human intervention, representing a significant step toward more autonomous and adaptable robotic systems. Demonstration videos are available at https://adaptive-intelligent-robotics.github.io/URSA.
Authors: Rivaaj Monsia, Olivier Francon, Daniel Young, Risto Miikkulainen
Abstract: This short, written report introduces the idea of Evolutionary Surrogate-Assisted Prescription (ESP) and presents preliminary results on its potential use in training real-world agents as a part of the 1st AI for Drinking Water Chlorination Challenge at IJCAI-2025. This work was done by a team from Project Resilience, an organization interested in bridging AI to real-world problems.
Authors: Alex LaGrassa, Zixuan Huang, Dmitry Berenson, Oliver Kroemer
Abstract: Efficient planning in high-dimensional spaces, such as those involving deformable objects, requires computationally tractable yet sufficiently expressive dynamics models. This paper introduces a method that automatically generates task-specific, spatially adaptive dynamics models by learning which regions of the object require high-resolution modeling to achieve good task performance for a given planning query. Task performance depends on the complex interplay between the dynamics model, world dynamics, control, and task requirements. Our proposed diffusion-based model generator predicts per-region model resolutions based on start and goal pointclouds that define the planning query. To efficiently collect the data for learning this mapping, a two-stage process optimizes resolution using predictive dynamics as a prior before directly optimizing using closed-loop performance. On a tree-manipulation task, our method doubles planning speed with only a small decrease in task performance over using a full-resolution model. This approach informs a path towards using previous planning and control data to generate computationally efficient yet sufficiently expressive dynamics models for new tasks.
Authors: Rudy Bunel, Jingyue Lu, Ilker Turkaslan, Philip H. S. Torr, Pushmeet Kohli, M. Pawan Kumar
Abstract: The success of Deep Learning and its potential use in many safety-critical applications has motivated research on formal verification of Neural Network (NN) models. In this context, verification involves proving or disproving that an NN model satisfies certain input-output properties. Despite the reputation of learned NN models as black boxes, and the theoretical hardness of proving useful properties about them, researchers have been successful in verifying some classes of models by exploiting their piecewise linear structure and taking insights from formal methods such as Satisifiability Modulo Theory. However, these methods are still far from scaling to realistic neural networks. To facilitate progress on this crucial area, we exploit the Mixed Integer Linear Programming (MIP) formulation of verification to propose a family of algorithms based on Branch-and-Bound (BaB). We show that our family contains previous verification methods as special cases. With the help of the BaB framework, we make three key contributions. Firstly, we identify new methods that combine the strengths of multiple existing approaches, accomplishing significant performance improvements over previous state of the art. Secondly, we introduce an effective branching strategy on ReLU non-linearities. This branching strategy allows us to efficiently and successfully deal with high input dimensional problems with convolutional network architecture, on which previous methods fail frequently. Finally, we propose comprehensive test data sets and benchmarks which includes a collection of previously released testcases. We use the data sets to conduct a thorough experimental comparison of existing and new algorithms and to provide an inclusive analysis of the factors impacting the hardness of verification problems.
Authors: Xin Du, Subramanian Ramamoorthy, Wouter Duivesteijn, Jin Tian, Mykola Pechenizkiy
Abstract: Local decision rules are commonly understood to be more explainable, due to the local nature of the patterns involved. With numerical optimization methods such as gradient boosting, ensembles of local decision rules can gain good predictive performance on data involving global structure. Meanwhile, machine learning models are being increasingly used to solve problems in high-stake domains including healthcare and finance. Here, there is an emerging consensus regarding the need for practitioners to understand whether and how those models could perform robustly in the deployment environments, in the presence of distributional shifts. Past research on local decision rules has focused mainly on maximizing discriminant patterns, without due consideration of robustness against distributional shifts. In order to fill this gap, we propose a new method to learn and ensemble local decision rules, that are robust both in the training and deployment environments. Specifically, we propose to leverage causal knowledge by regarding the distributional shifts in subpopulations and deployment environments as the results of interventions on the underlying system. We propose two regularization terms based on causal knowledge to search for optimal and stable rules. Experiments on both synthetic and benchmark datasets show that our method is effective and robust against distributional shifts in multiple environments.
Authors: Jonathan W. Siegel
Abstract: We study the interpolation power of deep ReLU neural networks. Specifically, we consider the question of how efficiently, in terms of the number of parameters, deep ReLU networks can interpolate values at $N$ datapoints in the unit ball which are separated by a distance $\delta$. We show that $\Omega(N)$ parameters are required in the regime where $\delta$ is exponentially small in $N$, which gives the sharp result in this regime since $O(N)$ parameters are always sufficient. This also shows that the bit-extraction technique used to prove lower bounds on the VC dimension cannot be applied to irregularly spaced datapoints. Finally, as an application we give a lower bound on the approximation rates that deep ReLU neural networks can achieve for Sobolev spaces at the embedding endpoint.
Authors: Tianyu Wang, Jiashuo Liu, Peng Cui, Hongseok Namkoong
Abstract: Different distribution shifts require different interventions, and algorithms must be grounded in the specific shifts they address. However, methodological development for robust algorithms typically relies on structural assumptions that lack empirical validation. Advocating for an empirically grounded data-driven approach to algorithm development, we build an empirical testbed comprising natural shifts across 8 tabular datasets, 172 distribution pairs over 45 methods and 90,000 method configurations encompassing empirical risk minimization and distributionally robust optimization (DRO) methods. We find $Y|X$-shifts are most prevalent in our testbed, in stark contrast to the heavy focus on $X$ (covariate)-shifts in the ML literature, and that the performance of robust algorithms is no better than that of vanilla methods. To understand why, we conduct an in-depth empirical analysis of DRO methods and find that underlooked implementation details -- such as the choice of underlying model class (e.g., LightGBM) and hyperparameter selection -- have a bigger impact on performance than the ambiguity set or its radius. We illustrate via case studies how a data-driven, inductive understanding of distribution shifts can provide a new approach to algorithm development.
Authors: Michaela Ennis, Leo Kozachkov, Jean-Jacques Slotine
Abstract: To push forward the important emerging research field surrounding multi-area recurrent neural networks (RNNs), we expand theoretically and empirically on the provably stable RNNs of RNNs introduced by Kozachkov et al. in "RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks". We prove relaxed stability conditions for salient special cases of this architecture, most notably for a global workspace modular structure. We then demonstrate empirical success for Global Workspace Sparse Combo Nets with a small number of trainable parameters, not only through strong overall test performance but also greater resilience to removal of individual subnetworks. These empirical results for the global workspace inter-area topology are contingent on stability preservation, highlighting the relevance of our theoretical work for enabling modular RNN success. Further, by exploring sparsity in the connectivity structure between different subnetwork modules more broadly, we improve the state of the art performance for stable RNNs on benchmark sequence processing tasks, thus underscoring the general utility of specialized graph structures for multi-area RNNs.
Authors: Nathan Justin, Sina Aghaei, Andr\'es G\'omez, Phebe Vayanos
Abstract: We consider the problem of learning classification trees that are robust to distribution shifts between training and testing/deployment data. This problem arises frequently in high stakes settings such as public health and social work where data is often collected using self-reported surveys which are highly sensitive to e.g., the framing of the questions, the time when and place where the survey is conducted, and the level of comfort the interviewee has in sharing information with the interviewer. We propose a method for learning optimal robust classification trees based on mixed-integer robust optimization technology. In particular, we demonstrate that the problem of learning an optimal robust tree can be cast as a single-stage mixed-integer robust optimization problem with a highly nonlinear and discontinuous objective. We reformulate this problem equivalently as a two-stage linear robust optimization problem for which we devise a tailored solution procedure based on constraint generation. We evaluate the performance of our approach on numerous publicly available datasets, and compare the performance to a regularized, non-robust optimal tree. We show an increase of up to 12.48% in worst-case accuracy and of up to 4.85% in average-case accuracy across several datasets and distribution shifts from using our robust solution in comparison to the non-robust one.
Authors: Mingkui Tan, Guohao Chen, Jiaxiang Wu, Yifan Zhang, Yaofo Chen, Peilin Zhao, Shuaicheng Niu
Abstract: Test-time adaptation (TTA) seeks to tackle potential distribution shifts between training and test data by adapting a given model w.r.t. any test sample. Although recent TTA has shown promising performance, we still face two key challenges: 1) prior methods perform backpropagation for each test sample, resulting in unbearable optimization costs to many applications; 2) while existing TTA can significantly improve the test performance on out-of-distribution data, they often suffer from severe performance degradation on in-distribution data after TTA (known as forgetting). To this end, we have proposed an Efficient Anti-Forgetting Test-Time Adaptation (EATA) method which develops an active sample selection criterion to identify reliable and non-redundant samples for test-time entropy minimization. To alleviate forgetting, EATA introduces a Fisher regularizer estimated from test samples to constrain important model parameters from drastic changes. However, in EATA, the adopted entropy loss consistently assigns higher confidence to predictions even for samples that are underlying uncertain, leading to overconfident predictions. To tackle this, we further propose EATA with Calibration (EATA-C) to separately exploit the reducible model uncertainty and the inherent data uncertainty for calibrated TTA. Specifically, we measure the model uncertainty by the divergence between predictions from the full network and its sub-networks, on which we propose a divergence loss to encourage consistent predictions instead of overconfident ones. To further recalibrate prediction confidence, we utilize the disagreement among predicted labels as an indicator of the data uncertainty, and then devise a min-max entropy regularizer to selectively increase and decrease prediction confidence for different samples. Experiments on image classification and semantic segmentation verify the effectiveness of our methods.
Authors: Taihei Oki, Shinsaku Sakaue
Abstract: M${}^{\natural}$-concave functions, a.k.a. gross substitute valuation functions, play a fundamental role in many fields, including discrete mathematics and economics. In practice, perfect knowledge of M${}^{\natural}$-concave functions is often unavailable a priori, and we can optimize them only interactively based on some feedback. Motivated by such situations, we study online M${}^{\natural}$-concave function maximization problems, which are interactive versions of the problem studied by Murota and Shioura (1999). For the stochastic bandit setting, we present $O(T^{-1/2})$-simple regret and $O(T^{2/3})$-regret algorithms under $T$ times access to unbiased noisy value oracles of M${}^{\natural}$-concave functions. A key to proving these results is the robustness of the greedy algorithm to local errors in M${}^{\natural}$-concave function maximization, which is one of our main technical results. While we obtain those positive results for the stochastic setting, another main result of our work is an impossibility in the adversarial setting. We prove that, even with full-information feedback, no algorithms that run in polynomial time per round can achieve $O(T^{1-c})$ regret for any constant $c > 0$. Our proof is based on a reduction from the matroid intersection problem for three matroids, which would be a novel approach to establishing the hardness in online learning.
Authors: Lev Telyatnikov, Guillermo Bernardez, Marco Montagna, Mustafa Hajij, Martin Carrasco, Pavlo Vasylenko, Mathilde Papillon, Ghada Zamzmi, Michael T. Schaub, Jonas Verhellen, Pavel Snopov, Bertran Miquel-Oliver, Manel Gil-Sorribes, Alexis Molina, Victor Guallar, Theodore Long, Julian Suk, Patryk Rygiel, Alexander Nikitin, Giordan Escalona, Michael Banf, Dominik Filipiak, Max Schattauer, Liliya Imasheva, Alvaro Martinez, Halley Fritze, Marissa Masden, Valentina S\'anchez, Manuel Lecha, Andrea Cavallo, Claudio Battiloro, Matt Piekenbrock, Mauricio Tec, George Dasoulas, Nina Miolane, Simone Scardapane, Theodore Papamarkou
Abstract: This work introduces TopoBench, an open-source library designed to standardize benchmarking and accelerate research in topological deep learning (TDL). TopoBench decomposes TDL into a sequence of independent modules for data generation, loading, transforming and processing, as well as model training, optimization and evaluation. This modular organization provides flexibility for modifications and facilitates the adaptation and optimization of various TDL pipelines. A key feature of TopoBench is its support for transformations and lifting across topological domains. Mapping the topology and features of a graph to higher-order topological domains, such as simplicial and cell complexes, enables richer data representations and more fine-grained analyses. The applicability of TopoBench is demonstrated by benchmarking several TDL architectures across diverse tasks and datasets.
Authors: Jongwoo Kim, Seongyeub Chu, Hyeongmin Park, Bryan Wong, Keejun Han, Mun Yong Yi
Abstract: Recent advancements in graph neural networks (GNNs) and heterogeneous GNNs (HGNNs) have advanced node embeddings and relationship learning for various tasks. However, existing methods often rely on domain-specific predefined meta-paths, which are coarse-grained and focus solely on aspects like node type, limiting their ability to capture complex interactions. We introduce MF2Vec, a model that uses multi-faceted (fine-grained) paths instead of predefined meta-paths. MF2Vec extracts paths via random walks and generates multi-faceted vectors, ignoring predefined schemas. This method learns diverse aspects of nodes and their relationships, constructs a homogeneous network, and creates node embeddings for classification, link prediction, and clustering. Extensive experiments show that MF2Vec outperforms existing methods, offering a more flexible and comprehensive framework for analyzing complex networks. The code is available at https://anonymous.4open.science/r/MF2Vec-6ABC.
Authors: Huiying Liu, Zekun Zhang, Honghao Li, Qilin Wu, Yiwen Zhang
Abstract: Large language models (LLMs) have seen rapid improvement in the recent years, and have been used in a wider range of applications. After being trained on large text corpus, LLMs obtain the capability of extracting rich features from textual data. Such capability is potentially useful for the web service recommendation task, where the web users and services have intrinsic attributes that can be described using natural language sentences and are useful for recommendation. In this paper, we explore the possibility and practicality of using LLMs for web service recommendation. We propose the large language model aided QoS prediction (llmQoS) model, which use LLMs to extract useful information from attributes of web users and services via descriptive sentences. This information is then used in combination with the QoS values of historical interactions of users and services, to predict QoS values for any given user-service pair. On the WSDream dataset, llmQoS is shown to overcome the data sparsity issue inherent to the QoS prediction problem, and outperforms comparable baseline models consistently.
Authors: Bella Finkel, Jose Israel Rodriguez, Chenxi Wu, Thomas Yahl
Abstract: We study the expressive power of deep polynomial neural networks through the geometry of their neurovariety. We introduce the notion of the activation degree threshold of a network architecture to express when the dimension of the neurovariety achieves its theoretical maximum. We prove the existence of the activation degree threshold for all polynomial neural networks without width-one bottlenecks and demonstrate a universal upper bound that is quadratic in the width of largest size. In doing so, we prove the high activation degree conjecture of Kileel, Trager, and Bruna. Certain structured architectures have exceptional activation degree thresholds, making them especially expressive in the sense of their neurovariety dimension. In this direction, we prove that polynomial neural networks with equi-width architectures are maximally expressive by showing their activation degree threshold is one.
Authors: Yuran Xiang, Haiteng Zhao, Chang Ma, Zhi-Hong Deng
Abstract: Recent advancements in computational chemistry have increasingly focused on synthesizing molecules based on textual instructions. Integrating graph generation with these instructions is complex, leading most current methods to use molecular sequences with pre-trained large language models. In response to this challenge, we propose a novel framework, named $\textbf{UTGDiff (Unified Text-Graph Diffusion Model)}$, which utilizes language models for discrete graph diffusion to generate molecular graphs from instructions. UTGDiff features a unified text-graph transformer as the denoising network, derived from pre-trained language models and minimally modified to process graph data through attention bias. Our experimental results demonstrate that UTGDiff consistently outperforms sequence-based baselines in tasks involving instruction-based molecule generation and editing, achieving superior performance with fewer parameters given an equivalent level of pretraining corpus. Our code is availble at https://github.com/ran1812/UTGDiff.
Authors: Jason Matthews, Alex Bihlo
Abstract: In recent years the study of deep learning for solving differential equations has grown substantially. The use of physics-informed neural networks (PINNs) and deep operator networks (DeepONets) have emerged as two of the most useful approaches in approximating differential equation solutions using machine learning. Here, we introduce PinnDE, an open-source Python library for solving differential equations with both PINNs and DeepONets. We give a brief review of both PINNs and DeepONets, introduce PinnDE along with the structure and usage of the package, and present worked examples to show PinnDE's effectiveness in approximating solutions of systems of differential equations with both PINNs and DeepONets.
Authors: A. Yark{\i}n Y{\i}ld{\i}z, Asli Kalayci
Abstract: Medical diagnosis is a crucial task in the medical field, in terms of providing accurate classification and respective treatments. Having near-precise decisions based on correct diagnosis can affect a patient's life itself, and may extremely result in a catastrophe if not classified correctly. Several traditional machine learning (ML), such as support vector machines (SVMs) and logistic regression, and state-of-the-art tabular deep learning (DL) methods, including TabNet and TabTransformer, have been proposed and used over tabular medical datasets. Additionally, due to the superior performances, lower computational costs, and easier optimization over different tasks, ensemble methods have been used in the field more recently. They offer a powerful alternative in terms of providing successful medical decision-making processes in several diagnosis tasks. In this study, we investigated the benefits of ensemble methods, especially the Gradient Boosting Decision Tree (GBDT) algorithms in medical classification tasks over tabular data, focusing on XGBoost, CatBoost, and LightGBM. The experiments demonstrate that GBDT methods outperform traditional ML and deep neural network architectures and have the highest average rank over several benchmark tabular medical diagnosis datasets. Furthermore, they require much less computational power compared to DL models, creating the optimal methodology in terms of high performance and lower complexity.
Authors: Weijia Xu, Nebojsa Jojic, Nicolas Le Roux
Abstract: Humans can learn to solve new tasks by inducing high-level strategies from example solutions to similar problems and then adapting these strategies to solve unseen problems. Can we use large language models to induce such high-level structure from example documents or solutions? We introduce fLSA, a foundation-model-based Latent Semantic Analysis method that iteratively clusters and tags document segments based on document-level contexts. These tags can be used to model the latent structure of given documents and for hierarchical sampling of new texts. Our experiments on story writing, math, and multi-step reasoning datasets demonstrate that fLSA tags are more informative in reconstructing the original texts than existing tagging methods. Moreover, when used for hierarchical sampling, fLSA tags help expand the output space in the right directions that lead to correct solutions more often than direct sampling and hierarchical sampling with existing tagging methods. Code: https://github.com/microsoft/fLSA
Authors: Utsav Singh, Souradip Chakraborty, Wesley A. Suttle, Brian M. Sadler, Derrik E. Asher, Anit Kumar Sahu, Mubarak Shah, Vinay P. Namboodiri, Amrit Singh Bedi
Abstract: Hierarchical reinforcement learning (HRL) enables agents to solve complex, long-horizon tasks by decomposing them into manageable sub-tasks. However, HRL methods often suffer from two fundamental challenges: (i) non-stationarity, caused by the changing behavior of the lower-level policy during training, which destabilizes higher-level policy learning, and (ii) the generation of infeasible subgoals that lower-level policies cannot achieve. In this work, we introduce DIPPER, a novel HRL framework that formulates hierarchical policy learning as a bi-level optimization problem and leverages direct preference optimization (DPO) to train the higher-level policy using preference feedback. By optimizing the higher-level policy with DPO, we decouple higher-level learning from the non-stationary lower-level reward signal, thus mitigating non-stationarity. To further address the infeasible subgoal problem, DIPPER incorporates a regularization that tries to ensure the feasibility of subgoal tasks within the capabilities of the lower-level policy. Extensive experiments on challenging robotic navigation and manipulation benchmarks demonstrate that DIPPER achieves up to 40\% improvement over state-of-the-art baselines in sparse reward scenarios, highlighting its effectiveness in overcoming longstanding limitations of HRL.
Authors: Edvin Listo Zec, Adam Breitholtz, Fredrik D. Johansson
Abstract: Federated learning enables multiple actors to collaboratively train models without sharing private data. Existing algorithms are successful and well-justified in this task when the intended target domain, where the trained model will be used, shares data distribution with the aggregate of clients, but this is often violated in practice. A common reason is label shift -- that the label distributions differ between clients and the target domain. We demonstrate empirically that this can significantly degrade performance. To address this problem, we propose FedPALS, a principled and practical model aggregation scheme that adapts to label shifts to improve performance in the target domain by leveraging knowledge of label distributions at the central server. Our approach ensures unbiased updates under federated stochastic gradient descent which yields robust generalization across clients with diverse, label-shifted data. Extensive experiments on image classification tasks demonstrate that FedPALS consistently outperforms baselines by aligning model aggregation with the target domain. Our findings reveal that conventional federated learning methods suffer severely in cases of extreme label sparsity on clients, highlighting the critical need for target-aware aggregation as offered by FedPALS.
Authors: Levi Rauchwerger, Stefanie Jegelka, Ron Levie
Abstract: We analyze the universality and generalization of graph neural networks (GNNs) on attributed graphs, i.e., with node attributes. To this end, we propose pseudometrics over the space of all attributed graphs that describe the fine-grained expressivity of GNNs. Namely, GNNs are both Lipschitz continuous with respect to our pseudometrics and can separate attributed graphs that are distant in the metric. Moreover, we prove that the space of all attributed graphs is relatively compact with respect to our metrics. Based on these properties, we prove a universal approximation theorem for GNNs and generalization bounds for GNNs on any data distribution of attributed graphs. The proposed metrics compute the similarity between the structures of attributed graphs via a hierarchical optimal transport between computation trees. Our work extends and unites previous approaches which either derived theory only for graphs with no attributes, derived compact metrics under which GNNs are continuous but without separation power, or derived metrics under which GNNs are continuous and separate points but the space of graphs is not relatively compact, which prevents universal approximation and generalization analysis.
Authors: Shaojie Bai, Mohammad Sadegh Talebi, Chengcheng Zhao, Peng Cheng, Jiming Chen
Abstract: Reinforcement learning (RL) is a powerful tool for sequential decision-making, but its application is often hindered by privacy concerns arising from its interaction data. This challenge is particularly acute in advanced Cyber-Physical Systems (CPS), where learning from operational and user data can expose systems to privacy inference attacks. Existing differential privacy (DP) models for RL are often inadequate: the centralized model requires a fully trusted server, creating a single point of failure risk, while the local model incurs significant performance degradation that is unsuitable for many control applications. This paper addresses this gap by leveraging the emerging shuffle model of privacy, an intermediate trust model that provides strong privacy guarantees without a centralized trust assumption. We present Shuffle Differentially Private Policy Elimination (SDP-PE), the first generic policy elimination-based algorithm for episodic RL under the shuffle model. Our method introduces a novel exponential batching schedule and a ``forgetting'' mechanism to balance the competing demands of privacy and learning performance. Our analysis shows that SDP-PE achieves a near-optimal regret bound, demonstrating a superior privacy-regret trade-off that significantly outperforms the local model. This work establishes the viability of the shuffle model for secure data-driven control in advanced CPS.
Authors: Rajesh Mangannavar, Alan Fern, Prasad Tadepalli
Abstract: We present an online planning framework and a new benchmark dataset for solving multi-object rearrangement problems in partially observable, multi-room environments. Current object rearrangement solutions, primarily based on Reinforcement Learning or hand-coded planning methods, often lack adaptability to diverse challenges. To address this limitation, we introduce a novel Hierarchical Object-Oriented Partially Observed Markov Decision Process (HOO-POMDP) planning approach. This approach comprises of (a) an object-oriented POMDP planner generating sub-goals, (b) a set of low-level policies for sub-goal achievement, and (c) an abstraction system converting the continuous low-level world into a representation suitable for abstract planning. To enable rigorous evaluation of rearrangement challenges, we introduce MultiRoomR, a comprehensive benchmark featuring diverse multi-room environments with varying degrees of partial observability (10-30\% initial visibility), blocked paths, obstructed goals, and multiple objects (10-20) distributed across 2-4 rooms. Experiments demonstrate that our system effectively handles these complex scenarios while maintaining robust performance even with imperfect perception, achieving promising results across both existing benchmarks and our new MultiRoomR dataset.
Authors: Rajesh Mangannavar, Stefan Lee, Alan Fern, Prasad Tadepalli
Abstract: We propose a novel approach to learn relational policies for classical planning based on learning to rank actions. We introduce a new graph representation that explicitly captures action information and propose a Graph Neural Network (GNN) architecture augmented with Gated Recurrent Units (GRUs) to learn action rankings. Unlike value-function based approaches that must learn a globally consistent function, our action ranking method only needs to learn locally consistent ranking, which is more sample-efficient. Our model is trained on data generated from small problem instances that are easily solved by planners and is applied to significantly larger instances where planning is computationally prohibitive. Experimental results across standard planning benchmarks demonstrate that our action-ranking approach not only achieves better generalization to larger problems than those used in training but also outperforms multiple baseline (value function and action ranking) methods in terms of success rate and plan quality.
Authors: Long Kiu Chung, Shreyas Kousik
Abstract: Even though neural networks are being increasingly deployed in safety-critical control applications, it remains difficult to enforce constraints on their output, meaning that it is hard to guarantee safety in such settings. While many existing methods seek to verify a neural network's satisfaction of safety constraints, few address how to correct an unsafe network. The handful of works that extract a training signal from verification cannot handle non-convex sets, and are either conservative or slow. To begin addressing these challenges, this work proposes a neural network training method that can encourage the exact image of a non-convex input set for a neural network with rectified linear unit (ReLU) nonlinearities to avoid a non-convex unsafe region. This is accomplished by reachability analysis with scaled hybrid zonotopes, a modification of the existing hybrid zonotope set representation that enables parameterized scaling of non-convex polytopic sets with a differentiable collision check via mixed-integer linear programs (MILPs). The proposed method was shown to be effective and fast for networks with up to 240 neurons, with the computational complexity dominated by inverse operations on matrices that scale linearly in size with the number of neurons and complexity of input and unsafe sets. We demonstrate the practicality of our method by training a forward-invariant neural network controller for an affine dynamical system with a non-convex input set, as well as generating safe reach-avoid plans for a black-box dynamical system.
Authors: Dylan Cutler, Arun Kandoor, Nishanth Dikkala, Nikunj Saunshi, Xin Wang, Rina Panigrahy
Abstract: Decoding in a Transformer based language model is inherently sequential as a token's embedding needs to pass through all the layers in the network before the generation of the next token can begin. In this work, we propose a new architecture StagFormer (Staggered Transformer), which staggers execution along the sequence axis and thereby enables parallelizing the decoding process along the depth of the model. We achieve this by breaking the dependency of the token representation at time step $i$ in layer $l$ upon the representations of tokens until time step $i$ from layer $l-1$. Instead, we stagger the execution and only allow a dependency on token representations until time step $i-1$. The later sections of the Transformer still get access to the "rich" representations from the prior section but only from those token positions which are one time step behind. StagFormer allows for different sections of the model to be executed in parallel yielding a potential speedup in decoding while being quality neutral in our simulations. We also explore many natural extensions of this idea. We present how weight-sharing across the different sections being staggered can be more practical in settings with limited memory. We explore the efficacy of using a bounded window attention to pass information from one section to another which helps drive further latency gains for some applications. We also explore the scalability of the staggering idea over more than 2 sections of the Transformer. Finally, we show how one can approximate a recurrent model during inference using weight-sharing. This variant can lead to substantial gains in quality for short generations while being neutral in its latency impact.
Authors: Cassandra Mussard, Arthur Charpentier, St\'ephane Mussard
Abstract: This paper introduces enhancements to the K-means and K-nearest neighbors (KNN) algorithms based on the concept of Gini prametric spaces, instead of traditional metric spaces. Unlike standard distance metrics, Gini prametrics incorporate both value-based and rank-based measures, offering robustness to noise and outliers. The main contributions include: (1) a Gini prametric that captures rank information alongside value distances; (2) a Gini K-means algorithm that is provably convergent and resilient to noisy data; and (3) a Gini KNN method that performs competitively with state-of-the-art approaches like Hassanat's distance in noisy environments. Experimental evaluations on 16 UCI datasets demonstrate the superior performance and efficiency of the Gini-based algorithms in clustering and classification tasks. This work opens new directions for rank-based prametrics in machine learning and statistical analysis.
Authors: Evgeniia Tokarchuk, Hua Chang Bakker, Vlad Niculae
Abstract: Learning well-separated features in high-dimensional spaces, such as text or image embeddings, is crucial for many machine learning applications. Achieving such separation can be effectively accomplished through the dispersion of embeddings, where unrelated vectors are pushed apart as much as possible. By constraining features to be on a hypersphere, we can connect dispersion to well-studied problems in mathematics and physics, where optimal solutions are known for limited low-dimensional cases. However, in representation learning we typically deal with a large number of features in high-dimensional space, and moreover, dispersion is usually traded off with some other task-oriented training objective, making existing theoretical and numerical solutions inapplicable. Therefore, it is common to rely on gradient-based methods to encourage dispersion, usually by minimizing some function of the pairwise distances. In this work, we first give an overview of existing methods from disconnected literature, making new connections and highlighting similarities. Next, we introduce some new angles. We propose to reinterpret pairwise dispersion using a maximum mean discrepancy (MMD) motivation. We then propose an online variant of the celebrated Lloyd's algorithm, of K-Means fame, as an effective alternative regularizer for dispersion on generic domains. Finally, we derive a novel dispersion method that directly exploits properties of the hypersphere. Our experiments show the importance of dispersion in image classification and natural language processing tasks, and how algorithms exhibit different trade-offs in different regimes.
Authors: Seungwook Han, Jyothish Pari, Samuel J. Gershman, Pulkit Agrawal
Abstract: Large Language Models (LLMs) have demonstrated impressive real-world utility, exemplifying artificial useful intelligence (AUI). However, their ability to reason adaptively and robustly -- the hallmarks of artificial general intelligence (AGI) -- remains fragile. While LLMs seemingly succeed in commonsense reasoning, programming, and mathematics, they struggle to generalize algorithmic understanding across novel contexts. Our experiments with algorithmic tasks in esoteric programming languages reveal that LLM's reasoning overfits to the training data and is limited in its transferability. We hypothesize that the core issue underlying such limited transferability is the coupling of reasoning and knowledge in LLMs. To transition from AUI to AGI, we propose disentangling knowledge and reasoning through three key directions: (1) pretaining to reason using RL from scratch as an alternative to the widely used next-token prediction pretraining, (2) using a curriculum of synthetic tasks to ease the learning of a reasoning prior for RL that can then be transferred to natural language tasks, and (3) learning more generalizable reasoning functions using a small context window to reduce exploiting spurious correlations between tokens. Such a reasoning system coupled with a trained retrieval system and a large external memory bank as a knowledge store can overcome several limitations of existing architectures at learning to reason in novel scenarios.
Authors: Gongbo Zhang, Yanting Li, Renqian Luo, Pipi Hu, Yang Yang, Zeru Zhao, Lingbo Li, Guoqing Liu, Zun Wang, Ran Bi, Kaiyuan Gao, Liya Guo, Yu Xie, Chang Liu, Jia Zhang, Tian Xie, Robert Pinsler, Claudio Zeni, Ziheng Lu, Hongxia Hao, Yingce Xia, Marwin Segler, Maik Riechert, Wei Yang, Hao Jiang, Wen-Bin Zhang, Zhijun Zeng, Yi Zhu, Li Dong, Xiuyuan Hu, Li Yuan, Lei Chen, Haiguang Liu, Tao Qin
Abstract: Function in natural systems arises from one-dimensional sequences forming three-dimensional structures with specific properties. However, current generative models suffer from critical limitations: training objectives seldom target function directly, discrete sequences and continuous coordinates are optimized in isolation, and conformational ensembles are under-modeled. We present UniGenX, a unified generative foundation model that addresses these gaps by co-generating sequences and coordinates under direct functional and property objectives across proteins, molecules, and materials. UniGenX represents heterogeneous inputs as a mixed stream of symbolic and numeric tokens, where a decoder-only autoregressive transformer provides global context and a conditional diffusion head generates numeric fields steered by task-specific tokens. Besides the new high SOTAs on structure prediction tasks, the model demonstrates state-of-the-art or competitive performance for the function-aware generation across domains: in materials, it achieves "conflicted" multi-property conditional generation, yielding 436 crystal candidates meeting triple constraints, including 11 with novel compositions; in chemistry, it sets new benchmarks on five property targets and conformer ensemble generation on GEOM; and in biology, it improves success in modeling protein induced fit (RMSD < 2 {\AA}) by over 23-fold and enhances EC-conditioned enzyme design. Ablation studies and cross-domain transfer substantiate the benefits of joint discrete-continuous training, establishing UniGenX as a significant advance from prediction to controllable, function-aware generation.
Authors: Ivan Saboli\'c, Matej Grci\'c, Sini\v{s}a \v{S}egvi\'c
Abstract: We propose VIBE, a model-agnostic framework that trains classifiers resilient to backdoor attacks. The key concept behind our approach is to treat malicious inputs and corrupted labels from the training dataset as observed random variables, while the actual clean labels are latent. VIBE then recovers the corresponding latent clean label posterior through variational inference. The resulting training procedure follows the expectation-maximization (EM) algorithm. The E-step infers the clean pseudolabels by solving an entropy-regularized optimal transport problem, while the M-step updates the classifier parameters via gradient descent. Being modular, VIBE can seamlessly integrate with recent advancements in self-supervised representation learning, which enhance its ability to resist backdoor attacks. We experimentally validate the method effectiveness against contemporary backdoor attacks on standard datasets, a large-scale setup with 1$k$ classes, and a dataset poisoned with multiple attacks. VIBE consistently outperforms previous defenses across all tested scenarios.
Authors: Jes\'us Garc\'ia Fern\'andez, Nasir Ahmad, Marcel van Gerven
Abstract: Biological neural systems efficiently learn from delayed rewards despite relying on noisy synaptic transmission and lacking centralized optimization mechanisms. In contrast, artificial neural networks trained with reinforcement learning typically rely on backpropagation (BP), which limits their use in resource-constrained systems or with non-differentiable components. While noise-based alternatives, like reward-modulated Hebbian learning (RMHL), provide a biologically grounded framework for credit assignment, they struggle with temporal delays and hierarchical processing -key challenges in real-world learning. In this work, we derive a novel noise-based learning rule to address these challenges. Drawing inspiration from biological neural circuits, our method uses reward prediction errors as its optimization target to generate increasingly advantageous behavior, and incorporates an eligibility trace to facilitate retrospective credit assignment. Its formulation relies on local information, aligning with biological constraints and enabling neuromorphic implementation. Experimental validation on reinforcement tasks (immediate and delayed rewards) shows our approach significantly outperforms RMHL and achieves performance comparable to BP, although with slower convergence due to its noise-driven updates. While tested on simple architectures, the results highlight the potential of noise-driven, brain-inspired learning for low-power adaptive systems, particularly in scenarios where energy efficiency and biological plausibility are a priority. These findings also offer mechanistic insights into how dopamine-like signals and synaptic stochasticity may jointly enable learning in biological networks, bridging computational models with neurobiological principles.
Authors: Junkyum Kim, Divya Mahajan
Abstract: Retrieval-Augmented Generation (RAG) systems combine vector similarity search with large language models (LLMs) to deliver accurate, context-aware responses. However, co-locating the vector retriever and the LLM on shared GPU infrastructure introduces significant challenges: vector search is memory and I/O intensive, while LLM inference demands high throughput and low latency. Naive resource sharing often leads to severe performance degradation, particularly under high request load or large index sizes. We present VectorLiteRAG, a deployment-friendly RAG system that achieves latency-compliant inference without requiring additional hardware resources. VectorLiteRAG introduces a fine-grained GPU resource allocation mechanism based on detailed performance modeling and access pattern analysis. By estimating search latency and query hit rate distributions, it identifies an optimal index partitioning point across CPU and GPU tiers to minimize contention and maximize throughput. Our evaluations show that VectorLiteRAG consistently expands the SLO compliant request rate range across all tested configurations, including both small and large LLMs, and small and large vector databases compared to naive baselines and state of the art alternatives. In the best case, VectorLiteRAG improves the attainable SLO throughput by up to 1.5 times without compromising generation quality or requiring additional compute resources.
Authors: Benjamin C. Koenig, Suyong Kim, Sili Deng
Abstract: Efficient chemical kinetic model inference and application in combustion are challenging due to large ODE systems and widely separated time scales. Machine learning techniques have been proposed to streamline these models, though strong nonlinearity and numerical stiffness combined with noisy data sources make their application challenging. Here, we introduce ChemKANs, a novel neural network framework with applications both in model inference and simulation acceleration for combustion chemistry. ChemKAN's novel structure augments the generic Kolmogorov Arnold Network Ordinary Differential Equations (KAN-ODEs) with knowledge of the information flow through the relevant kinetic and thermodynamic laws. This chemistry-specific structure combined with the expressivity and rapid neural scaling of the underlying KAN-ODE algorithm instills in ChemKANs a strong inductive bias, streamlined training, and higher accuracy predictions compared to standard benchmarks, while facilitating parameter sparsity through shared information across all inputs and outputs. In a model inference investigation, we benchmark the robustness of ChemKANs to sparse data containing up to 15% added noise, and superfluously large network parameterizations. We find that ChemKANs exhibit no overfitting or model degradation in any of these training cases, demonstrating significant resilience to common deep learning failure modes. Next, we find that a remarkably parameter-lean ChemKAN (344 parameters) can accurately represent hydrogen combustion chemistry, providing a 2x acceleration over the detailed chemistry in a solver that is generalizable to larger-scale turbulent flow simulations. These demonstrations indicate the potential for ChemKANs as robust, expressive, and efficient tools for model inference and simulation acceleration for combustion physics and chemical kinetics.
Authors: Shuchen Wu, Stephan Alaniz, Shyamgopal Karthik, Peter Dayan, Eric Schulz, Zeynep Akata
Abstract: Neural networks are often described as black boxes, reflecting the significant challenge of understanding their internal workings and interactions. We propose a different perspective that challenges the prevailing view: rather than being inscrutable, neural networks exhibit patterns in their raw population activity that mirror regularities in the training data. We refer to this as the Reflection Hypothesis and provide evidence for this phenomenon in both simple recurrent neural networks (RNNs) and complex large language models (LLMs). Building on this insight, we propose to leverage our cognitive tendency of chunking to segment high-dimensional neural population dynamics into interpretable units that reflect underlying concepts. We propose three methods to extract recurring chunks on a neural population level, complementing each other based on label availability and neural data dimensionality. Discrete sequence chunking (DSC) learns a dictionary of entities in a lower-dimensional neural space; population averaging (PA) extracts recurring entities that correspond to known labels; and unsupervised chunk discovery (UCD) can be used when labels are absent. We demonstrate the effectiveness of these methods in extracting concept-encoding entities agnostic to model architectures. These concepts can be both concrete (words), abstract (POS tags), or structural (narrative schema). Additionally, we show that extracted chunks play a causal role in network behavior, as grafting them leads to controlled and predictable changes in the model's behavior. Our work points to a new direction for interpretability, one that harnesses both cognitive principles and the structure of naturalistic data to reveal the hidden computations of complex learning systems, gradually transforming them from black boxes into systems we can begin to understand.
Authors: Yuan Yuan, Yukun Liu, Chonghua Han, Jie Feng, Yong Li
Abstract: Foundation models have revolutionized fields such as natural language processing and computer vision by enabling general-purpose learning across diverse tasks and datasets. However, building analogous models for human mobility remains challenging due to the privacy-sensitive nature of mobility data and the resulting data silos across institutions. To bridge this gap, we propose MoveGCL, a scalable and privacy-preserving framework for training mobility foundation models via generative continual learning. Without sharing raw data, MoveGCL enables decentralized and progressive model evolution by replaying synthetic trajectories generated from a frozen teacher model, and reinforces knowledge retention through a tailored distillation strategy that mitigates catastrophic forgetting. To address the heterogeneity of mobility patterns, MoveGCL incorporates a Mixture-of-Experts Transformer with a mobility-aware expert routing mechanism, and employs a layer-wise progressive adaptation strategy to stabilize continual updates. Experiments on six real-world urban datasets demonstrate that MoveGCL achieves performance comparable to joint training and significantly outperforms federated learning baselines, while offering strong privacy protection. MoveGCL marks a crucial step toward unlocking foundation models for mobility, offering a practical blueprint for open, scalable, and privacy-preserving model development in the era of foundation models. To facilitate reproducibility and future research, we have released the code and models at https://github.com/tsinghua-fib-lab/MoveGCL.
Authors: Yufeng Wang, Peiyao Wang, Lu Wei, Lu Ma, Yuewei Lin, Qun Liu, Haibin Ling
Abstract: X-ray Absorption Spectroscopy (XAS) is a powerful technique for probing local atomic environments, yet its interpretation remains limited by the need for expert-driven analysis, computationally expensive simulations, and element-specific heuristics. Recent advances in machine learning have shown promise for accelerating XAS interpretation, but many existing models are narrowly focused on specific elements, edge types, or spectral regimes. In this work, we present XAStruct, a learning-based system capable of both predicting XAS spectra from crystal structures and inferring local structural descriptors from XAS input. XAStruct is trained on a large-scale dataset spanning over 70 elements across the periodic table, enabling generalization to a wide variety of chemistries and bonding environments. The framework includes the first machine learning approach for predicting neighbor atom types directly from XAS spectra, as well as a generalizable regression model for mean nearest-neighbor distance that requires no element-specific tuning. By combining deep neural networks for complex structure property mappings with efficient baseline models for simpler tasks, XAStruct offers a scalable and extensible solution for data-driven XAS analysis and local structure inference. The source code will be released upon paper acceptance.
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 local learning rules. The gradient with respect to the parameters of the driving protocol is 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 12x12 oscillator network on the MNIST dataset to generate images of handwritten digits 0 and 1.
Authors: Fouad Oubari, Raphael Meunier, Rodrigue D\'ecatoire, Mathilde Mougeot
Abstract: As deep generative models proliferate across the AI landscape, industrial practitioners still face critical yet unanswered questions about which deep generative models best suit complex manufacturing design tasks. This work addresses this question through a complete study of five representative models (Variational Autoencoder, Generative Adversarial Network, multimodal Variational Autoencoder, Denoising Diffusion Probabilistic Model, and Multinomial Diffusion Model) on industrial tire architecture generation. Our evaluation spans three key industrial scenarios: (i) unconditional generation of complete multi-component designs, (ii) component-conditioned generation (reconstructing architectures from partial observations), and (iii) dimension-constrained generation (creating designs that satisfy specific dimensional requirements). To enable discrete diffusion models to handle conditional scenarios, we introduce categorical inpainting, a mask-aware reverse diffusion process that preserves known labels without requiring additional training. Our evaluation employs geometry-aware metrics specifically calibrated for industrial requirements, quantifying spatial coherence, component interaction, structural connectivity, and perceptual fidelity. Our findings reveal that diffusion models achieve the strongest overall performance; a masking-trained VAE nonetheless outperforms the multimodal variant MMVAE\textsuperscript{+} on nearly all component-conditioned metrics, and within the diffusion family MDM leads in-distribution whereas DDPM generalises better to out-of-distribution dimensional constraints.
Authors: Fouad Oubari, Mohamed El-Baha, Raphael Meunier, Rodrigue D\'ecatoire, Mathilde Mougeot
Abstract: Multi-component datasets with intricate dependencies, like industrial assemblies or multi-modal imaging, challenge current generative modeling techniques. Existing Multi-component Variational AutoEncoders typically rely on simplified aggregation strategies, neglecting critical nuances and consequently compromising structural coherence across generated components. To explicitly address this gap, we introduce the Gaussian Markov Random Field Multi-Component Variational AutoEncoder , a novel generative framework embedding Gaussian Markov Random Fields into both prior and posterior distributions. This design choice explicitly models cross-component relationships, enabling richer representation and faithful reproduction of complex interactions. Empirically, our GMRF MCVAE achieves state-of-the-art performance on a synthetic Copula dataset specifically constructed to evaluate intricate component relationships, demonstrates competitive results on the PolyMNIST benchmark, and significantly enhances structural coherence on the real-world BIKED dataset. Our results indicate that the GMRF MCVAE is especially suited for practical applications demanding robust and realistic modeling of multi-component coherence
Authors: Ethan Li (Taoyi), Anders Boesen Lindbo Larsen (Taoyi), Chen Zhang (Taoyi), Xiyou Zhou (Taoyi), Jun Qin (Taoyi), Dian Ang Yap (Taoyi), Narendran Raghavan (Taoyi), Xuankai Chang (Taoyi), Margit Bowler (Taoyi), Eray Yildiz (Taoyi), John Peebles (Taoyi), Hannah Gillis Coleman (Taoyi), Matteo Ronchi (Taoyi), Peter Gray (Taoyi), Keen You (Taoyi), Anthony Spalvieri-Kruse (Taoyi), Ruoming Pang (Taoyi), Reed Li (Taoyi), Yuli Yang (Taoyi), Emad Soroush (Taoyi), Zhiyun Lu (Taoyi), Crystal Xiao (Taoyi), Rong Situ (Taoyi), Jordan Huffaker (Taoyi), David Griffiths (Taoyi), Zaid Ahmed (Taoyi), Peng Zhang (Taoyi), Daniel Parilla (Taoyi), Asaf Liberman (Taoyi), Jennifer Mallalieu (Taoyi), Parsa Mazaheri (Taoyi), Qibin Chen (Taoyi), Manjot Bilkhu (Taoyi), Aonan Zhang (Taoyi), Eric Wang (Taoyi), Dave Nelson (Taoyi), Michael FitzMaurice (Taoyi), Thomas Voice (Taoyi), Jeremy Liu (Taoyi), Josh Shaffer (Taoyi), Shiwen Zhao (Taoyi), Prasanth Yadla (Taoyi), Farzin Rasteh (Taoyi), Pengsheng Guo (Taoyi), Arsalan Farooq (Taoyi), Jeremy Snow (Taoyi), Stephen Murphy (Taoyi), Tao Lei (Taoyi), Minsik Cho (Taoyi), George Horrell (Taoyi), Sam Dodge (Taoyi), Lindsay Hislop (Taoyi), Sumeet Singh (Taoyi), Alex Dombrowski (Taoyi), Aiswarya Raghavan (Taoyi), Sasha Sirovica (Taoyi), Mandana Saebi (Taoyi), Faye Lao (Taoyi), Max Lam (Taoyi), TJ Lu (Taoyi), Zhaoyang Xu (Taoyi), Karanjeet Singh (Taoyi), Marc Kirchner (Taoyi), David Mizrahi (Taoyi), Rajat Arora (Taoyi), Haotian Zhang (Taoyi), Henry Mason (Taoyi), Lawrence Zhou (Taoyi), Yi Hua (Taoyi), Ankur Jain (Taoyi), Felix Bai (Taoyi), Joseph Astrauskas (Taoyi), Floris Weers (Taoyi), Josh Gardner (Taoyi), Mira Chiang (Taoyi), Yi Zhang (Taoyi), Pulkit Agrawal (Taoyi), Tony Sun (Taoyi), Quentin Keunebroek (Taoyi), Matthew Hopkins (Taoyi), Bugu Wu (Taoyi), Tao Jia (Taoyi), Chen Chen (Taoyi), Xingyu Zhou (Taoyi), Nanzhu Wang (Taoyi), Peng Liu (Taoyi), Ruixuan Hou (Taoyi), Rene Rauch (Taoyi), Yuan Gao (Taoyi), Afshin Dehghan (Taoyi), Jonathan Janke (Taoyi), Zirui Wang (Taoyi), Cha Chen (Taoyi), Xiaoyi Ren (Taoyi), Feng Nan (Taoyi), Josh Elman (Taoyi), Dong Yin (Taoyi), Yusuf Goren (Taoyi), Jeff Lai (Taoyi), Yiran Fei (Taoyi), Syd Evans (Taoyi), Muyang Yu (Taoyi), Guoli Yin (Taoyi), Yi Qin (Taoyi), Erin Feldman (Taoyi), Isha Garg (Taoyi), Aparna Rajamani (Taoyi), Karla Vega (Taoyi), Walker Cheng (Taoyi), TJ Collins (Taoyi), Hans Han (Taoyi), Raul Rea Menacho (Taoyi), Simon Yeung (Taoyi), Sophy Lee (Taoyi), Phani Mutyala (Taoyi), Ying-Chang Cheng (Taoyi), Zhe Gan (Taoyi), Sprite Chu (Taoyi), Justin Lazarow (Taoyi), Alessandro Pappalardo (Taoyi), Federico Scozzafava (Taoyi), Jing Lu (Taoyi), Erik Daxberger (Taoyi), Laurent Duchesne (Taoyi), Jen Liu (Taoyi), David G\"uera (Taoyi), Stefano Ligas (Taoyi), Mary Beth Kery (Taoyi), Brent Ramerth (Taoyi), Ciro Sannino (Taoyi), Marcin Eichner (Taoyi), Haoshuo Huang (Taoyi), Rui Qian (Taoyi), Moritz Schwarzer-Becker (Taoyi), David Riazati (Taoyi), Mingfei Gao (Taoyi), Bailin Wang (Taoyi), Jack Cackler (Taoyi), Yang Lu (Taoyi), Ransen Niu (Taoyi), John Dennison (Taoyi), Guillaume Klein (Taoyi), Jeffrey Bigham (Taoyi), Deepak Gopinath (Taoyi), Navid Shiee (Taoyi), Darren Botten (Taoyi), Guillaume Tartavel (Taoyi), Alex Guillen Garcia (Taoyi), Sam Xu (Taoyi), Victoria M\"onchJuan Haladjian (Taoyi), Zi-Yi Dou (Taoyi), Matthias Paulik (Taoyi), Adolfo Lopez Mendez (Taoyi), Zhen Li (Taoyi), Hong-You Chen (Taoyi), Chao Jia (Taoyi), Dhaval Doshi (Taoyi), Zhengdong Zhang (Taoyi), Raunak Manjani (Taoyi), Aaron Franklin (Taoyi), Zhile Ren (Taoyi), David Chen (Taoyi), Artsiom Peshko (Taoyi), Nandhitha Raghuram (Taoyi), Hans Hao (Taoyi), Jiulong Shan (Taoyi), Kavya Nerella (Taoyi), Ramsey Tantawi (Taoyi), Vivek Kumar (Taoyi), Saiwen Wang (Taoyi), Brycen Wershing (Taoyi), Bhuwan Dhingra (Taoyi), Dhruti Shah (Taoyi), Ob Adaranijo (Taoyi), Xin Zheng (Taoyi), Tait Madsen (Taoyi), Hadas Kotek (Taoyi), Chang Liu (Taoyi), Yin Xia (Taoyi), Hanli Li (Taoyi), Suma Jayaram (Taoyi), Yanchao Sun (Taoyi), Ahmed Fakhry (Taoyi), Vasileios Saveris (Taoyi), Dustin Withers (Taoyi), Yanghao Li (Taoyi), Alp Aygar (Taoyi), Andres Romero Mier Y Teran (Taoyi), Kaiwei Huang (Taoyi), Mark Lee (Taoyi), Xiujun Li (Taoyi), Yuhong Li (Taoyi), Tyler Johnson (Taoyi), Jay Tang (Taoyi), Joseph Yitan Cheng (Taoyi), Futang Peng (Taoyi), Andrew Walkingshaw (Taoyi), Lucas Guibert (Taoyi), Abhishek Sharma (Taoyi), Cheng Shen (Taoyi), Piotr Maj (Taoyi), Yasutaka Tanaka (Taoyi), You-Cyuan Jhang (Taoyi), Vivian Ma (Taoyi), Tommi Vehvilainen (Taoyi), Kelvin Zou (Taoyi), Jeff Nichols (Taoyi), Matthew Lei (Taoyi), David Qiu (Taoyi), Yihao Qian (Taoyi), Gokul Santhanam (Taoyi), Wentao Wu (Taoyi), Yena Han (Taoyi), Dominik Moritz (Taoyi), Haijing Fu (Taoyi), Mingze Xu (Taoyi), Vivek Rathod (Taoyi), Jian Liu (Taoyi), Louis D'hauwe (Taoyi), Qin Ba (Taoyi), Haitian Sun (Taoyi), Haoran Yan (Taoyi), Philipp Dufter (Taoyi), Anh Nguyen (Taoyi), Yihao Feng (Taoyi), Emma Wang (Taoyi), Keyu He (Taoyi), Rahul Nair (Taoyi), Sanskruti Shah (Taoyi), Jiarui Lu (Taoyi), Patrick Sonnenberg (Taoyi), Jeremy Warner (Taoyi), Yuanzhi Li (Taoyi), Bowen Pan (Taoyi), Ziyi Zhong (Taoyi), Joe Zhou (Taoyi), Sam Davarnia (Taoyi), Olli Saarikivi (Taoyi), Irina Belousova (Taoyi), Rachel Burger (Taoyi), Shang-Chen Wu (Taoyi), Di Feng (Taoyi), Bas Straathof (Taoyi), James Chou (Taoyi), Yuanyang Zhang (Taoyi), Marco Zuliani (Taoyi), Eduardo Jimenez (Taoyi), Abhishek Sundararajan (Taoyi), Xianzhi Du (Taoyi), Chang Lan (Taoyi), Nilesh Shahdadpuri (Taoyi), Peter Grasch (Taoyi), Sergiu Sima (Taoyi), Josh Newnham (Taoyi), Varsha Paidi (Taoyi), Jianyu Wang (Taoyi), Kaelen Haag (Taoyi), Alex Braunstein (Taoyi), Daniele Molinari (Taoyi), Richard Wei (Taoyi), Brenda Yang (Taoyi), Nicholas Lusskin (Taoyi), Joanna Arreaza-Taylor (Taoyi), Meng Cao (Taoyi), Nicholas Seidl (Taoyi), Simon Wang (Taoyi), Jiaming Hu (Taoyi), Yiping Ma (Taoyi), Mengyu Li (Taoyi), Kieran Liu (Taoyi), Hang Su (Taoyi), Sachin Ravi (Taoyi), Chong Wang (Taoyi), Xin Wang (Taoyi), Kevin Smith (Taoyi), Haoxuan You (Taoyi), Binazir Karimzadeh (Taoyi), Rui Li (Taoyi), Jinhao Lei (Taoyi), Wei Fang (Taoyi), Alec Doane (Taoyi), Sam Wiseman (Taoyi), Ismael Fernandez (Taoyi), Jane Li (Taoyi), Andrew Hansen (Taoyi), Javier Movellan (Taoyi), Christopher Neubauer (Taoyi), Hanzhi Zhou (Taoyi), Chris Chaney (Taoyi), Nazir Kamaldin (Taoyi), Valentin Wolf (Taoyi), Fernando Berm\'udez-Medina (Taoyi), Joris Pelemans (Taoyi), Peter Fu (Taoyi), Howard Xing (Taoyi), Xiang Kong (Taoyi), Wayne Shan (Taoyi), Gabriel Jacoby-Cooper (Taoyi), Dongcai Shen (Taoyi), Tom Gunter (Taoyi), Guillaume Seguin (Taoyi), Fangping Shi (Taoyi), Shiyu Li (Taoyi), Yang Xu (Taoyi), Areeba Kamal (Taoyi), Dan Masi (Taoyi), Saptarshi Guha (Taoyi), Qi Zhu (Taoyi), Jenna Thibodeau (Taoyi), Changyuan Zhang (Taoyi), Rebecca Callahan (Taoyi), Charles Maalouf (Taoyi), Wilson Tsao (Taoyi), Boyue Li (Taoyi), Qingqing Cao (Taoyi), Naomy Sabo (Taoyi), Cheng Leong (Taoyi), Yi Wang (Taoyi), Anupama Mann Anupama (Taoyi), Colorado Reed (Taoyi), Kenneth Jung (Taoyi), Zhifeng Chen (Taoyi), Mohana Prasad Sathya Moorthy (Taoyi), Yifei He (Taoyi), Erik Hornberger (Taoyi), Devi Krishna (Taoyi), Senyu Tong (Taoyi), Michael (Taoyi), Lee, David Haldimann, Yang Zhao, Bowen Zhang, Chang Gao, Chris Bartels, Sushma Rao, Nathalie Tran, Simon Lehnerer, Co Giang, Patrick Dong, Junting Pan, Biyao Wang, Dongxu Li, Mehrdad Farajtabar, Dongseong Hwang, Grace Duanmu, Eshan Verma, Sujeeth Reddy, Qi Shan, Hongbin Gao, Nan Du, Pragnya Sridhar, Forrest Huang, Yingbo Wang, Nikhil Bhendawade, Diane Zhu, Sai Aitharaju, Fred Hohman, Lauren Gardiner, Chung-Cheng Chiu, Yinfei Yang, Alper Kokmen, Frank Chu, Ke Ye, Kaan Elgin, Oron Levy, John Park, Donald Zhang, Eldon Schoop, Nina Wenzel, Michael Booker, Hyunjik Kim, Chinguun Erdenebileg, Nan Dun, Eric Liang Yang, Priyal Chhatrapati, Vishaal Mahtani, Haiming Gang, Kohen Chia, Deepa Seshadri, Donghan Yu, Yan Meng, Kelsey Peterson, Zhen Yang, Yongqiang Wang, Carina Peng, Doug Kang, Anuva Agarwal, Albert Antony, Juan Lao Tebar, Albin Madappally Jose, Regan Poston, Andy De Wang, Gerard Casamayor, Elmira Amirloo, Violet Yao, Wojciech Kryscinski, Kun Duan, Lezhi L
Abstract: We introduce two multilingual, multimodal foundation language models that power Apple Intelligence features across Apple devices and services: i a 3B-parameter on-device model optimized for Apple silicon through architectural innovations such as KV-cache sharing and 2-bit quantization-aware training; and ii a scalable server model built on a novel Parallel-Track Mixture-of-Experts PT-MoE transformer that combines track parallelism, mixture-of-experts sparse computation, and interleaved global-local attention to deliver high quality with competitive cost on Apple's Private Cloud Compute platform. Both models are trained on large-scale multilingual and multimodal datasets sourced via responsible web crawling, licensed corpora, and high-quality synthetic data, then further refined with supervised fine-tuning and reinforcement learning on a new asynchronous platform. The resulting models support several additional languages while understanding images and executing tool calls. In public benchmarks and human evaluations, both the server model and the on-device model match or surpass comparably sized open baselines. A new Swift-centric Foundation Models framework exposes guided generation, constrained tool calling, and LoRA adapter fine-tuning, allowing developers to integrate these capabilities with a few lines of code. The latest advancements in Apple Intelligence models are grounded in our Responsible AI approach with safeguards like content filtering and locale-specific evaluation, as well as our commitment to protecting our users' privacy with innovations like Private Cloud Compute.
Authors: Shiko Kudo
Abstract: The dominant paradigm in modern neural networks relies on simple, monotonically-increasing activation functions like ReLU. While effective, this paradigm necessitates large, massively-parameterized models to approximate complex functions. In this paper, we introduce the Periodic Linear Unit (PLU), a learnable sine-wave based activation with periodic non-monotonicity. PLU is designed for maximum expressive power and numerical stability, achieved through its formulation and a paired innovation we term Repulsive Reparameterization, which prevents the activation from collapsing into a non-expressive linear function. We demonstrate that a minimal MLP with only two PLU neurons can solve the spiral classification task, a feat impossible for equivalent networks using standard activations. This suggests a paradigm shift from networks as piecewise Taylor-like approximators to powerful Fourier-like function synthesizers, achieving exponential gains in parameter efficiency by placing intelligence in the neuron itself.
Authors: Ze Tao, Hanxuan Wang, Fujun Liu
Abstract: Physics-informed neural networks (PINNs) have attracted considerable attention for their ability to integrate partial differential equation priors into deep learning frameworks; however, they often exhibit limited predictive accuracy when applied to complex problems. To address this issue, we propose LNN-PINN, a physics-informed neural network framework that incorporates a liquid residual gating architecture while preserving the original physics modeling and optimization pipeline to improve predictive accuracy. The method introduces a lightweight gating mechanism solely within the hidden-layer mapping, keeping the sampling strategy, loss composition, and hyperparameter settings unchanged to ensure that improvements arise purely from architectural refinement. Across four benchmark problems, LNN-PINN consistently reduced RMSE and MAE under identical training conditions, with absolute error plots further confirming its accuracy gains. Moreover, the framework demonstrates strong adaptability and stability across varying dimensions, boundary conditions, and operator characteristics. In summary, LNN-PINN offers a concise and effective architectural enhancement for improving the predictive accuracy of physics-informed neural networks in complex scientific and engineering problems.
Authors: Bilal Faye, Hanane Azzag, Mustapha Lebbah
Abstract: Diffusion models have emerged as a leading framework for high-quality image generation, offering stable training and strong performance across diverse domains. However, they remain computationally intensive, particularly during the iterative denoising process. Latent-space models like Stable Diffusion alleviate some of this cost by operating in compressed representations, though at the expense of fine-grained detail. More recent approaches such as Retrieval-Augmented Diffusion Models (RDM) address efficiency by conditioning denoising on similar examples retrieved from large external memory banks. While effective, these methods introduce drawbacks: they require costly storage and retrieval infrastructure, depend on static vision-language models like CLIP for similarity, and lack adaptability during training. We propose the Prototype Diffusion Model (PDM), a method that integrates prototype learning directly into the diffusion process for efficient and adaptive visual conditioning - without external memory. Instead of retrieving reference samples, PDM constructs a dynamic set of compact visual prototypes from clean image features using contrastive learning. These prototypes guide the denoising steps by aligning noisy representations with semantically relevant visual patterns, enabling efficient generation with strong semantic grounding. Experiments show that PDM maintains high generation quality while reducing computational and storage overhead, offering a scalable alternative to retrieval-based conditioning in diffusion models.
Authors: Thore Wietzke, Knut Graichen
Abstract: Gaussian Processes (GPs) are widely used for regression and system identification due to their flexibility and ability to quantify uncertainty. However, their computational complexity limits their applicability to small datasets. Moreover in a streaming scenario, more and more datapoints accumulate which is intractable even for Sparse GPs. Online GPs aim to alleviate this problem by e.g. defining a maximum budget of datapoints and removing redundant datapoints. This work provides a unified comparison of several reduction criteria, analyzing both their computational complexity and reduction behavior. The criteria are evaluated on benchmark functions and real-world datasets, including dynamic system identification tasks. Additionally, acceptance criteria are proposed to further filter out redundant datapoints. This work yields practical guidelines for choosing a suitable criterion for an online GP algorithm.
Authors: Max Guillen, Philipp Misof, Jan E. Gerken
Abstract: Neural tangent kernels (NTKs) are a powerful tool for analyzing deep, non-linear neural networks. In the infinite-width limit, NTKs can easily be computed for most common architectures, yielding full analytic control over the training dynamics. However, at infinite width, important properties of training such as NTK evolution or feature learning are absent. Nevertheless, finite width effects can be included by computing corrections to the Gaussian statistics at infinite width. We introduce Feynman diagrams for computing finite-width corrections to NTK statistics. These dramatically simplify the necessary algebraic manipulations and enable the computation of layer-wise recursive relations for arbitrary statistics involving preactivations, NTKs and certain higher-derivative tensors (dNTK and ddNTK) required to predict the training dynamics at leading order. We demonstrate the feasibility of our framework by extending stability results for deep networks from preactivations to NTKs and proving the absence of finite-width corrections for scale-invariant nonlinearities such as ReLU on the diagonal of the Gram matrix of the NTK. We validate our results with numerical experiments.
Authors: Chongyu Qu, Allen J. Luna, Thomas Z. Li, Junchao Zhu, Junlin Guo, Juming Xiong, Kim L. Sandler, Bennett A. Landman, Yuankai Huo
Abstract: Accurate lung cancer risk prediction remains challenging due to substantial variability across patient populations and clinical settings -- no single model performs best for all cohorts. To address this, we propose a personalized lung cancer risk prediction agent that dynamically selects the most appropriate model for each patient by combining cohort-specific knowledge with modern retrieval and reasoning techniques. Given a patient's CT scan and structured metadata -- including demographic, clinical, and nodule-level features -- the agent first performs cohort retrieval using FAISS-based similarity search across nine diverse real-world cohorts to identify the most relevant patient population from a multi-institutional database. Second, a Large Language Model (LLM) is prompted with the retrieved cohort and its associated performance metrics to recommend the optimal prediction algorithm from a pool of eight representative models, including classical linear risk models (e.g., Mayo, Brock), temporally-aware models (e.g., TD-VIT, DLSTM), and multi-modal computer vision-based approaches (e.g., Liao, Sybil, DLS, DLI). This two-stage agent pipeline -- retrieval via FAISS and reasoning via LLM -- enables dynamic, cohort-aware risk prediction personalized to each patient's profile. Building on this architecture, the agent supports flexible and cohort-driven model selection across diverse clinical populations, offering a practical path toward individualized risk assessment in real-world lung cancer screening.
Authors: Luca Zhou, Daniele Solombrino, Donato Crisostomi, Maria Sofia Bucarelli, Giuseppe Alessio D'Inverno, Fabrizio Silvestri, Emanuele Rodol\`a
Abstract: Task arithmetic has emerged as a simple yet powerful technique for model merging, enabling the combination of multiple finetuned models into one. Despite its empirical success, a clear theoretical explanation of why and when it works is lacking. This paper provides a rigorous theoretical foundation for task arithmetic by establishing a connection between task vectors and gradients of the task losses. We show that under standard gradient descent, a task vector generated from one epoch of finetuning is exactly equivalent to the negative gradient of the loss, scaled by the learning rate. For the practical multi-epoch setting, we prove that this equivalence holds approximately, with a second-order error term that we explicitly bound for feed-forward networks. Our empirical analysis across seven vision benchmarks corroborates our theory, demonstrating that the first-epoch gradient dominates the finetuning trajectory in both norm and direction. A key implication is that merging models finetuned for only a single epoch often yields performance comparable to merging fully converged models. These findings reframe task arithmetic as a form of approximate multitask learning, providing a clear rationale for its effectiveness and highlighting the critical role of early training dynamics in model merging.
Authors: Jongyeop Hyun, Bumsoo Kim
Abstract: Recent advancements in Large Language Models (LLMs) have significantly improved reasoning capabilities, with in-context learning (ICL) emerging as a key technique for adaptation without retraining. While previous works have focused on leveraging correct examples, recent research highlights the importance of learning from errors to enhance performance. However, existing methods lack a structured framework for analyzing and mitigating errors, particularly in Multimodal Large Language Models (MLLMs), where integrating visual and textual inputs adds complexity. To address this issue, we propose REFINE: Retrieval-Enhanced Feedback via In-context Neural Error-book, a teacher-student framework that systematically structures errors and provides targeted feedback. REFINE introduces three systematic queries to construct structured feedback -- Feed-Target, Feed-Check, and Feed-Path -- to enhance multimodal reasoning by prioritizing relevant visual information, diagnosing critical failure points, and formulating corrective actions. Unlike prior approaches that rely on redundant retrievals, REFINE optimizes structured feedback retrieval, improving inference efficiency, token usage, and scalability. Our results demonstrate substantial speedup, reduced computational costs, and successful generalization, highlighting REFINE's potential for enhancing multimodal reasoning.
Authors: Muchammad Daniyal Kautsar, Afra Majida Hariono, Widyawan, Syukron Abu Ishaq Alfarozi, Kuntpong Woraratpanya
Abstract: Large Language Models (LLMs) present significant deployment challenges due to their immense size and computational requirements. Model compression techniques are essential for making these models practical for resource-constrained environments. A prominent compression strategy is low-rank factorization via Singular Value Decomposition (SVD) to reduce model parameters by approximating weight matrices. However, standard SVD focuses on minimizing matrix reconstruction error, often leading to a substantial loss of the model's functional performance. This performance degradation occurs because existing methods do not adequately correct for the functional information lost during compression. To address this gap, we introduce Corrective Adaptive Low-Rank Decomposition (CALR), a two-component compression approach. CALR combines a primary path of SVD-compressed layers with a parallel, learnable, low-rank corrective module that is explicitly trained to recover the functional residual error. Our experimental evaluation on SmolLM2-135M, Qwen3-0.6B, and Llama-3.2-1B, demonstrates that CALR can reduce parameter counts by 26.93% to 51.77% while retaining 59.45% to 90.42% of the original model's performance, consistently outperforming LaCo, ShortGPT, and LoSparse. CALR's success shows that treating functional information loss as a learnable signal is a highly effective compression paradigm. This approach enables the creation of significantly smaller, more efficient LLMs, advancing their accessibility and practical deployment in real-world applications.
Authors: Yang Zhou, Sunzhu Li, Shunyu Liu, Wenkai Fang, Jiale Zhao, Jingwen Yang, Jianwei Lv, Kongcheng Zhang, Yihe Zhou, Hengtong Lu, Wei Chen, Yan Xie, Mingli Song
Abstract: Recent advances in Large Language Models (LLMs) have underscored the potential of Reinforcement Learning (RL) to facilitate the emergence of reasoning capabilities. Despite the encouraging results, a fundamental dilemma persists as RL improvement relies on learning from high-quality samples, yet the exploration for such samples remains bounded by the inherent limitations of LLMs. This, in effect, creates an undesirable cycle in which what cannot be explored cannot be learned. In this work, we propose Rubric-Scaffolded Reinforcement Learning (RuscaRL), a novel instructional scaffolding framework designed to break the exploration bottleneck for general LLM reasoning. Specifically, RuscaRL introduces checklist-style rubrics as (1) explicit scaffolding for exploration during rollout generation, where different rubrics are provided as external guidance within task instructions to steer diverse high-quality responses. This guidance is gradually decayed over time, encouraging the model to internalize the underlying reasoning patterns; (2) verifiable rewards for exploitation during model training, where we can obtain robust LLM-as-a-Judge scores using rubrics as references, enabling effective RL on general reasoning tasks. Extensive experiments demonstrate the superiority of the proposed RuscaRL across various benchmarks, effectively expanding reasoning boundaries under the best-of-N evaluation. Notably, RuscaRL significantly boosts Qwen2.5-7B-Instruct from 23.6 to 50.3 on HealthBench-500, surpassing GPT-4.1. Furthermore, our fine-tuned variant on Qwen3-30B-A3B-Instruct achieves 61.1 on HealthBench-500, outperforming leading LLMs including OpenAI-o3. This work is still in progress, and we will release the code, the models, and the datasets soon.
Authors: Xiaomeng Fan, Yuwei Wu, Zhi Gao, Mehrtash Harandi, Yunde Jia
Abstract: Hyperbolic neural networks (HNNs) have demonstrated notable efficacy in representing real-world data with hierarchical structures via exploiting the geometric properties of hyperbolic spaces characterized by negative curvatures. Curvature plays a crucial role in optimizing HNNs. Inappropriate curvatures may cause HNNs to converge to suboptimal parameters, degrading overall performance. So far, the theoretical foundation of the effect of curvatures on HNNs has not been developed. In this paper, we derive a PAC-Bayesian generalization bound of HNNs, highlighting the role of curvatures in the generalization of HNNs via their effect on the smoothness of the loss landscape. Driven by the derived bound, we propose a sharpness-aware curvature learning method to smooth the loss landscape, thereby improving the generalization of HNNs. In our method, we design a scope sharpness measure for curvatures, which is minimized through a bi-level optimization process. Then, we introduce an implicit differentiation algorithm that efficiently solves the bi-level optimization by approximating gradients of curvatures. We present the approximation error and convergence analyses of the proposed method, showing that the approximation error is upper-bounded, and the proposed method can converge by bounding gradients of HNNs. Experiments on four settings: classification, learning from long-tailed data, learning from noisy data, and few-shot learning show that our method can improve the performance of HNNs.
Authors: An Ning, Tai Yue Li, Nan Yow Chen
Abstract: We propose the Quantum Graph Attention Network (QGAT), a hybrid graph neural network that integrates variational quantum circuits into the attention mechanism. At its core, QGAT employs strongly entangling quantum circuits with amplitude-encoded node features to enable expressive nonlinear interactions. Distinct from classical multi-head attention that separately computes each head, QGAT leverages a single quantum circuit to simultaneously generate multiple attention coefficients. This quantum parallelism facilitates parameter sharing across heads, substantially reducing computational overhead and model complexity. Classical projection weights and quantum circuit parameters are optimized jointly in an end-to-end manner, ensuring flexible adaptation to learning tasks. Empirical results demonstrate QGAT's effectiveness in capturing complex structural dependencies and improved generalization in inductive scenarios, highlighting its potential for scalable quantum-enhanced learning across domains such as chemistry, biology, and network analysis. Furthermore, experiments confirm that quantum embedding enhances robustness against feature and structural noise, suggesting advantages in handling real-world noisy data. The modularity of QGAT also ensures straightforward integration into existing architectures, allowing it to easily augment classical attention-based models.
Authors: Robert Yang
Abstract: Bold claims about AI's role in science-from "AGI will cure all diseases" to promises of radically accelerated discovery-raise a central epistemic question: do large language models (LLMs) truly generate new knowledge, or do they merely remix memorized fragments? We propose unlearning-as-ablation as a falsifiable probe of constructive scientific discovery. The idea is to systematically remove a target result together with its forget-closure (supporting lemmas, paraphrases, and multi-hop entailments) and then evaluate whether the model can re-derive the result from only permitted axioms and tools. Success would indicate generative capability beyond recall; failure would expose current limits. Unlike prevailing motivations for unlearning-privacy, copyright, or safety-our framing repositions it as an epistemic probe for AI-for-Science. We outline a minimal pilot in mathematics and algorithms to illustrate feasibility, and sketch how the same approach could later be extended to domains such as physics or chemistry. This is a position paper: our contribution is conceptual and methodological, not empirical. We aim to stimulate discussion on how principled ablation tests could help distinguish models that reconstruct knowledge from those that merely retrieve it, and how such probes might guide the next generation of AI-for-Science benchmarks.
Authors: Dan Wang, Feng Jiang, Zhanquan Wang
Abstract: Accurate air quality prediction is becoming increasingly important in the environmental field. To address issues such as low prediction accuracy and slow real-time updates in existing models, which lead to lagging prediction results, we propose a Transformer-based spatiotemporal data prediction method (Ada-TransGNN) that integrates global spatial semantics and temporal behavior. The model constructs an efficient and collaborative spatiotemporal block set comprising a multi-head attention mechanism and a graph convolutional network to extract dynamically changing spatiotemporal dependency features from complex air quality monitoring data. Considering the interaction relationships between different monitoring points, we propose an adaptive graph structure learning module, which combines spatiotemporal dependency features in a data-driven manner to learn the optimal graph structure, thereby more accurately capturing the spatial relationships between monitoring points. Additionally, we design an auxiliary task learning module that enhances the decoding capability of temporal relationships by integrating spatial context information into the optimal graph structure representation, effectively improving the accuracy of prediction results. We conducted comprehensive evaluations on a benchmark dataset and a novel dataset (Mete-air). The results demonstrate that our model outperforms existing state-of-the-art prediction models in short-term and long-term predictions.
Authors: Jianhao Huang, Qunsong Zeng, Hongyang Du, Kaibin Huang
Abstract: Semantic communication (SemCom) has emerged as a promising paradigm for achieving unprecedented communication efficiency in sixth-generation (6G) networks by leveraging artificial intelligence (AI) to extract and transmit the underlying meanings of source data. However, deploying SemCom over digital systems presents new challenges, particularly in ensuring robustness against transmission errors that may distort semantically critical content. To address this issue, this paper proposes a novel framework, termed generative feature imputing, which comprises three key techniques. First, we introduce a spatial error concentration packetization strategy that spatially concentrates feature distortions by encoding feature elements based on their channel mappings, a property crucial for both the effectiveness and reduced complexity of the subsequent techniques. Second, building on this strategy, we propose a generative feature imputing method that utilizes a diffusion model to efficiently reconstruct missing features caused by packet losses. Finally, we develop a semantic-aware power allocation scheme that enables unequal error protection by allocating transmission power according to the semantic importance of each packet. Experimental results demonstrate that the proposed framework outperforms conventional approaches, such as Deep Joint Source-Channel Coding (DJSCC) and JPEG2000, under block fading conditions, achieving higher semantic accuracy and lower Learned Perceptual Image Patch Similarity (LPIPS) scores.
Authors: Weida Wang, Dongchen Huang, Jiatong Li, Tengchao Yang, Ziyang Zheng, Di Zhang, Dong Han, Benteng Chen, Binzhao Luo, Zhiyu Liu, Kunling Liu, Zhiyuan Gao, Shiqi Geng, Wei Ma, Jiaming Su, Xin Li, Shuchen Pu, Yuhan Shui, Qianjia Cheng, Zhihao Dou, Dongfei Cui, Changyong He, Jin Zeng, Zeke Xie, Mao Su, Dongzhan Zhou, Yuqiang Li, Wanli Ouyang, Yunqi Cai, Xi Dai, Shufei Zhang, Lei Bai, Jinguang Cheng, Zhong Fang, Hongming Weng
Abstract: We introduce CMPhysBench, designed to assess the proficiency of Large Language Models (LLMs) in Condensed Matter Physics, as a novel Benchmark. CMPhysBench is composed of more than 520 graduate-level meticulously curated questions covering both representative subfields and foundational theoretical frameworks of condensed matter physics, such as magnetism, superconductivity, strongly correlated systems, etc. To ensure a deep understanding of the problem-solving process,we focus exclusively on calculation problems, requiring LLMs to independently generate comprehensive solutions. Meanwhile, leveraging tree-based representations of expressions, we introduce the Scalable Expression Edit Distance (SEED) score, which provides fine-grained (non-binary) partial credit and yields a more accurate assessment of similarity between prediction and ground-truth. Our results show that even the best models, Grok-4, reach only 36 average SEED score and 28% accuracy on CMPhysBench, underscoring a significant capability gap, especially for this practical and frontier domain relative to traditional physics. The code anddataset are publicly available at https://github.com/CMPhysBench/CMPhysBench.
Authors: Kento Tomita, Katherine A. Skinner, Koki Ho
Abstract: Hazard detection is critical for enabling autonomous landing on planetary surfaces. Current state-of-the-art methods leverage traditional computer vision approaches to automate the identification of safe terrain from input digital elevation models (DEMs). However, performance for these methods can degrade for input DEMs with increased sensor noise. In the last decade, deep learning techniques have been developed for various applications. Nevertheless, their applicability to safety-critical space missions has often been limited due to concerns regarding their outputs' reliability. In response to these limitations, this paper proposes an application of the Bayesian deep-learning segmentation method for hazard detection. The developed approach enables reliable, safe landing site detection by: (i) generating simultaneously a safety prediction map and its uncertainty map via Bayesian deep learning and semantic segmentation; and (ii) using the uncertainty map to filter out the uncertain pixels in the prediction map so that the safe site identification is performed only based on the certain pixels (i.e., pixels for which the model is certain about its safety prediction). Experiments are presented with simulated data based on a Mars HiRISE digital terrain model by varying uncertainty threshold and noise levels to demonstrate the performance of the proposed approach.
Authors: Kwonyoung Kim, Jungin Park, Jiyoung Lee, Dongbo Min, Kwanghoon Sohn
Abstract: Online stereo adaptation tackles the domain shift problem, caused by different environments between synthetic (training) and real (test) datasets, to promptly adapt stereo models in dynamic real-world applications such as autonomous driving. However, previous methods often fail to counteract particular regions related to dynamic objects with more severe environmental changes. To mitigate this issue, we propose to incorporate an auxiliary point-selective network into a meta-learning framework, called PointFix, to provide a robust initialization of stereo models for online stereo adaptation. In a nutshell, our auxiliary network learns to fix local variants intensively by effectively back-propagating local information through the meta-gradient for the robust initialization of the baseline model. This network is model-agnostic, so can be used in any kind of architectures in a plug-and-play manner. We conduct extensive experiments to verify the effectiveness of our method under three adaptation settings such as short-, mid-, and long-term sequences. Experimental results show that the proper initialization of the base stereo model by the auxiliary network enables our learning paradigm to achieve state-of-the-art performance at inference.
Authors: Sungnyun Kim, Junsoo Lee, Kibeom Hong, Daesik Kim, Namhyuk Ahn
Abstract: In this study, we aim to enhance the capabilities of diffusion-based text-to-image (T2I) generation models by integrating diverse modalities beyond textual descriptions within a unified framework. To this end, we categorize widely used conditional inputs into three modality types: structure, layout, and attribute. We propose a multimodal T2I diffusion model, which is capable of processing all three modalities within a single architecture without modifying the parameters of the pre-trained diffusion model, as only a small subset of components is updated. Our approach sets new benchmarks in multimodal generation through extensive quantitative and qualitative comparisons with existing conditional generation methods. We demonstrate that DiffBlender effectively integrates multiple sources of information and supports diverse applications in detailed image synthesis. The code and demo are available at https://github.com/sungnyun/diffblender.
Authors: Daniel Bethell, Simos Gerasimou, Radu Calinescu, Calum Imrie
Abstract: Empowering safe exploration of reinforcement learning (RL) agents during training is a critical challenge towards their deployment in many real-world scenarios. When prior knowledge of the domain or task is unavailable, training RL agents in unknown, black-box environments presents an even greater safety risk. We introduce ADVICE (Adaptive Shielding with a Contrastive Autoencoder), a novel post-shielding technique that distinguishes safe and unsafe features of state-action pairs during training, and uses this knowledge to protect the RL agent from executing actions that yield likely hazardous outcomes. Our comprehensive experimental evaluation against state-of-the-art safe RL exploration techniques shows that ADVICE significantly reduces safety violations (approx 50%) during training, with a competitive outcome reward compared to other techniques.
Authors: Maris F. L. Galesloot, Marnix Suilen, Thiago D. Sim\~ao, Steven Carr, Matthijs T. J. Spaan, Ufuk Topcu, Nils Jansen
Abstract: Robust POMDPs extend classical POMDPs to incorporate model uncertainty using so-called uncertainty sets on the transition and observation functions, effectively defining ranges of probabilities. Policies for robust POMDPs must be (1) memory-based to account for partial observability and (2) robust against model uncertainty to account for the worst-case probability instances from the uncertainty sets. To compute such robust memory-based policies, we propose the pessimistic iterative planning (PIP) framework, which alternates between (1) selecting pessimistic POMDPs via worst-case probability instances from the uncertainty sets, and (2) computing finite-state controllers (FSCs) for these pessimistic POMDPs. Within PIP, we propose the rFSCNet algorithm, which optimizes a recurrent neural network to compute the FSCs. The empirical evaluation shows that rFSCNet can compute better-performing robust policies than several baselines and a state-of-the-art robust POMDP solver.
Authors: Zhenhua Xu, Qichen Liu, Zhebo Wang, Wenpeng Xing, Dezhang Kong, Mohan Li, Meng Han
Abstract: Backdoor-based fingerprinting has emerged as an effective technique for tracing the ownership of large language models. However, in real-world deployment scenarios, developers often instantiate multiple downstream models from a shared base model, and applying fingerprinting to each variant individually incurs prohibitive computational overhead. While inheritance-based approaches -- where fingerprints are embedded into the base model and expected to persist through fine-tuning -- appear attractive, they suffer from three key limitations: late-stage fingerprinting, fingerprint instability, and interference with downstream adaptation. To address these challenges, we propose a novel mechanism called the Fingerprint Vector. Our method first embeds a fingerprint into the base model via backdoor-based fine-tuning, then extracts a task-specific parameter delta as a fingerprint vector by computing the difference between the fingerprinted and clean models. This vector can be directly added to any structurally compatible downstream model, allowing the fingerprint to be transferred post hoc without additional fine-tuning. Extensive experiments show that Fingerprint Vector achieves comparable or superior performance to direct injection across key desiderata. It maintains strong effectiveness across diverse model architectures as well as mainstream downstream variants within the same family. It also preserves harmlessness and robustness in most cases. Even when slight robustness degradation is observed, the impact remains within acceptable bounds and is outweighed by the scalability benefits of our approach.
Authors: Michael Gnewuch, Kumar Harsha, Marcin Wnuk
Abstract: The mean squared error and regularized versions of it are standard loss functions in supervised machine learning. However, calculating these losses for large data sets can be computationally demanding. Modifying an approach of J. Dick and M. Feischl [Journal of Complexity 67 (2021)], we present algorithms to reduce extensive data sets to a smaller size using rank-1 lattices. Rank-1 lattices are quasi-Monte Carlo (QMC) point sets that are, if carefully chosen, well-distributed in a multidimensional unit cube. The compression strategy in the preprocessing step assigns every lattice point a pair of weights depending on the original data and responses, representing its relative importance. As a result, the compressed data makes iterative loss calculations in optimization steps much faster. We analyze the errors of our QMC data compression algorithms and the cost of the preprocessing step for functions whose Fourier coefficients decay sufficiently fast so that they lie in certain Wiener algebras or Korobov spaces. In particular, we prove that our approach can lead to arbitrary high convergence rates as long as the functions are sufficiently smooth.
Authors: Julian Suk, Guido Nannini, Patryk Rygiel, Christoph Brune, Gianluca Pontone, Alberto Redaelli, Jelmer M. Wolterink
Abstract: Cardiovascular hemodynamic fields provide valuable medical decision markers for coronary artery disease. Computational fluid dynamics (CFD) is the gold standard for accurate, non-invasive evaluation of these quantities in silico. In this work, we propose a time-efficient surrogate model, powered by machine learning, for the estimation of pulsatile hemodynamics based on steady-state priors. We introduce deep vectorised operators, a modelling framework for discretisation-independent learning on infinite-dimensional function spaces. The underlying neural architecture is a neural field conditioned on hemodynamic boundary conditions. Importantly, we show how relaxing the requirement of point-wise action to permutation-equivariance leads to a family of models that can be parametrised by message passing and self-attention layers. We evaluate our approach on a dataset of 74 stenotic coronary arteries extracted from coronary computed tomography angiography (CCTA) with patient-specific pulsatile CFD simulations as ground truth. We show that our model produces accurate estimates of the pulsatile velocity and pressure (approximation disparity 0.368 $\pm$ 0.079) while being agnostic ($p < 0.05$ in a one-way ANOVA test) to re-sampling of the source domain, i.e. discretisation-independent. This shows that deep vectorised operators are a powerful modelling tool for cardiovascular hemodynamics estimation in coronary arteries and beyond.
Authors: Peng Zhu, Yuante Li, Yifan Hu, Sheng Xiang, Qinyuan Liu, Dawei Cheng, Yuqi Liang
Abstract: As financial markets grow increasingly complex in the big data era, accurate stock prediction has become more critical. Traditional time series models, such as GRUs, have been widely used but often struggle to capture the intricate nonlinear dynamics of markets, particularly in the flexible selection and effective utilization of key historical information. Recently, methods like Graph Neural Networks and Reinforcement Learning have shown promise in stock prediction but require high data quality and quantity, and they tend to exhibit instability when dealing with data sparsity and noise. Moreover, the training and inference processes for these models are typically complex and computationally expensive, limiting their broad deployment in practical applications. Existing approaches also generally struggle to capture unobservable latent market states effectively, such as market sentiment and expectations, microstructural factors, and participant behavior patterns, leading to an inadequate understanding of market dynamics and subsequently impact prediction accuracy. To address these challenges, this paper proposes a stock prediction model, MCI-GRU, based on a multi-head cross-attention mechanism and an improved GRU. First, we enhance the GRU model by replacing the reset gate with an attention mechanism, thereby increasing the model's flexibility in selecting and utilizing historical information. Second, we design a multi-head cross-attention mechanism for learning unobservable latent market state representations, which are further enriched through interactions with both temporal features and cross-sectional features. Finally, extensive experiments on four main stock markets show that the proposed method outperforms SOTA techniques across multiple metrics. Additionally, its successful application in real-world fund management operations confirms its effectiveness and practicality.
Authors: Volodymyr Karpenko, Taimoor Tariq, Jorge Condor, Piotr Didyk
Abstract: Modern deep-learning super-resolution (SR) techniques process images and videos independently of the underlying content and viewing conditions. However, the sensitivity of the human visual system (HVS) to image details changes depending on the underlying image characteristics, such as spatial frequency, luminance, color, contrast, or motion; as well viewing condition aspects such as ambient lighting and distance to the display. This observation suggests that computational resources spent on up-sampling images/videos may be wasted whenever a viewer cannot resolve the synthesized details i.e the resolution of details exceeds the resolving capability of human vision. Motivated by this observation, we propose a human vision inspired and architecture-agnostic approach for controlling SR techniques to deliver visually optimal results while limiting computational complexity. Its core is an explicit Human Visual Processing Framework (HVPF) that dynamically and locally guides SR methods according to human sensitivity to specific image details and viewing conditions. We demonstrate the application of our framework in combination with network branching to improve the computational efficiency of SR methods. Quantitative and qualitative evaluations, including user studies, demonstrate the effectiveness of our approach in reducing FLOPS by factors of 2$\times$ and greater, without sacrificing perceived quality.
Authors: Prajwal Singh, Ashish Tiwari, Gautam Vashishtha, Shanmuganathan Raman
Abstract: Neural radiance fields (NeRF) have revolutionized photorealistic rendering of novel views for 3D scenes. Despite their growing popularity and efficiency as 3D resources, NeRFs face scalability challenges due to the need for separate models per scene and the cumulative increase in training time for multiple scenes. The potential for incrementally encoding multiple 3D scenes into a single NeRF model remains largely unexplored. To address this, we introduce Continual-Neural Graphics Primitives (C-NGP), a novel continual learning framework that integrates multiple scenes incrementally into a single neural radiance field. Using a generative replay approach, C-NGP adapts to new scenes without requiring access to old data. We demonstrate that C-NGP can accommodate multiple scenes without increasing the parameter count, producing high-quality novel-view renderings on synthetic and real datasets. Notably, C-NGP models all $8$ scenes from the Real-LLFF dataset together, with only a $2.2\%$ drop in PSNR compared to vanilla NeRF, which models each scene independently. Further, C-NGP allows multiple style edits in the same network.
Authors: Mahdi Rezapour
Abstract: Recommendation systems have become the fundamental services to facilitate users information access. Generally, recommendation system works by filtering historical behaviors to understand and learn users preferences. With the growth of online information, recommendations have become of crucial importance in information filtering to prevent the information overload problem. In this study, we considered hybrid post-fusion of two approaches of collaborative filtering, by using sequences of watched movies and considering the related movies rating. After considering both techniques and applying the weights matrix, the recommendations would be modified to correspond to the users preference as needed. We discussed that various weights would be set based on use cases. For instance, in cases where we have the rating for most classes, we will assign a higher weight to the rating matrix and in case where the rating is unavailable for the majority of cases, the higher weights might be assigned to the sequential dataset. An extensive discussion is made in the context of this paper. Sequential type of the watched movies was used in conjunction of the rating as especially that model might be inadequate in distinguishing users long-term preference and that does not account for the rating of the watched movies and thus that model along might not suffice. Extensive discussion was made regarding the literature and methodological approach to solve the problem.
Authors: Dimitrios Mallis, Ahmet Serdar Karadeniz, Sebastian Cavada, Danila Rukhovich, Niki Foteinopoulou, Kseniya Cherenkova, Anis Kacem, Djamila Aouada
Abstract: We propose CAD-Assistant, a general-purpose CAD agent for AI-assisted design. Our approach is based on a powerful Vision and Large Language Model (VLLM) as a planner and a tool-augmentation paradigm using CAD-specific tools. CAD-Assistant addresses multimodal user queries by generating actions that are iteratively executed on a Python interpreter equipped with the FreeCAD software, accessed via its Python API. Our framework is able to assess the impact of generated CAD commands on geometry and adapts subsequent actions based on the evolving state of the CAD design. We consider a wide range of CAD-specific tools including a sketch image parameterizer, rendering modules, a 2D cross-section generator, and other specialized routines. CAD-Assistant is evaluated on multiple CAD benchmarks, where it outperforms VLLM baselines and supervised task-specific methods. Beyond existing benchmarks, we qualitatively demonstrate the potential of tool-augmented VLLMs as general-purpose CAD solvers across diverse workflows.
Authors: Guangxiang Zhao, Saier Hu, Xiaoqi Jian, Jinzhu Wu, Yuhan Wu, Change Jia, Lin Sun, Xiangzheng Zhang
Abstract: In this paper, we propose a ``Generalization Stress Test" to assess Large Language Models' (LLMs) generalization ability under slight and controlled perturbations, including option length, problem types, and irrelevant noun replacements. We achieve novel and significant findings that, despite high benchmark scores, LLMs exhibit severe accuracy drops and unexpected biases (e.g., preference for longer distractors) when faced with these minor but content-preserving modifications. For example, Qwen 2.5 1.5B's MMLU score rises from 60 to 89 and drops from 89 to 36 when option lengths are changed without altering the question. Even GPT4o experiences a 25-point accuracy loss when problem types are changed, with a 6-point drop across all three modification categories. These analyses suggest that LLMs rely heavily on superficial cues rather than forming robust, abstract representations that generalize across formats, lexical variations, and irrelevant content shifts.
Authors: Xudong Lu, Haohao Gao, Renshou Wu, Shuai Ren, Xiaoxin Chen, Hongsheng Li, Fangyuan Li
Abstract: Large Language Models (LLMs) have become integral to daily life, especially advancing as intelligent assistants through on-device deployment on smartphones. However, existing LLM evaluation benchmarks predominantly focus on objective tasks like mathematics and coding in English, which do not necessarily reflect the practical use cases of on-device LLMs in real-world mobile scenarios, especially for Chinese users. To address these gaps, we introduce SmartBench, the first benchmark designed to evaluate the capabilities of on-device LLMs in Chinese mobile contexts. We analyze functionalities provided by representative smartphone manufacturers and divide them into five categories: text summarization, text Q&A, information extraction, content creation, and notification management, further detailed into 20 specific tasks. For each task, we construct high-quality datasets comprising 50 to 200 question-answer pairs that reflect everyday mobile interactions, and we develop automated evaluation criteria tailored for these tasks. We conduct comprehensive evaluations of on-device LLMs and MLLMs using SmartBench and also assess their performance after quantized deployment on real smartphone NPUs. Our contributions provide a standardized framework for evaluating on-device LLMs in Chinese, promoting further development and optimization in this critical area. Code and data will be available at https://github.com/vivo-ai-lab/SmartBench.
Authors: Shun-Cai Zhao, Yi-Meng Huang, Yi-Fan Yang, Zi-Ran Zhao
Abstract: Machine learning simulations of open quantum dynamics often rely on recursive predictors that accumulate error. We develop a non-recursive convolutional neural networks (CNNs) that maps system parameters and a redundant time encoding directly to excitation-energy-transfer populations in the Fenna-Matthews-Olson complex. The encoding-modified logistic plus $\tanh$ functions-normalizes time and resolves fast, transitional, and quasi-steady regimes, while physics-informed labels enforce population conservation and inter-site consistency. Trained only on $0\sim 7 ps$ reference trajectories generated with a Lindblad model in QuTiP, the network accurately predicts $0\sim100 ps$ dynamics across a range of reorganization energies, bath rates, and temperatures. Beyond $20 ps$, the absolute relative error remains below 0.05, demonstrating stable long-time extrapolation. By avoiding step-by-step recursion, the method suppresses error accumulation and generalizes across timescales. These results show that redundant time encoding enables data-efficient inference of long-time quantum dissipative dynamics in realistic pigment-protein complexes, and may aid the data-driven design of light-harvesting materials.
Authors: Anas Barakat, Wayne Lin, John Lazarsfeld, Antonios Varvitsiotis
Abstract: We introduce symmetric cone games (SCGs), a broad class of multi-player games where each player's strategy lies in a generalized simplex (the trace-one slice of a symmetric cone). This framework unifies a wide spectrum of settings, including normal-form games (simplex strategies), quantum games (density matrices), and continuous games with ball-constrained strategies. It also captures several structured machine learning and optimization problems, such as distance metric learning and Fermat-Weber facility location, as two-player zero-sum SCGs. To compute approximate Nash equilibria in two-player zero-sum SCGs, we propose a single online learning algorithm: Optimistic Symmetric Cone Multiplicative Weights Updates (OSCMWU). Unlike prior methods tailored to specific geometries, OSCMWU provides closed-form, projection-free updates over any symmetric cone and achieves an optimal $\tilde{\mathcal{O}}(1/\epsilon)$ iteration complexity for computing $\epsilon$-saddle points. Our analysis builds on the Optimistic Follow-the-Regularized-Leader framework and hinges on a key technical contribution: We prove that the symmetric cone negative entropy is strongly convex with respect to the trace-one norm. This result extends known results for the simplex and spectraplex to all symmetric cones, and may be of independent interest.
Authors: Asma Yamani, Nadeen AlAmoudi, Salma Albilali, Malak Baslyman, Jameleddine Hassine
Abstract: Machine Learning (ML) has been integrated into various software and systems. Two main components are essential for training an ML model: the training data and the ML algorithm. Given the critical role of data in ML system development, it has become increasingly important to assess the quality of data attributes and ensure that the data meets specific requirements before its utilization. This work proposes an approach to guide non-experts in identifying data requirements for ML systems using goal modeling. In this approach, we first develop the Data Requirement Goal Model (DRGM) by surveying the white literature to identify and categorize the issues and challenges faced by data scientists and requirement engineers working on ML-related projects. An initial DRGM was built to accommodate common tasks that would generalize across projects. Then, based on insights from both white and gray literature, a customization mechanism is built to help adjust the tasks, KPIs, and goals' importance of different elements within the DRGM. The generated model can aid its users in evaluating different datasets using GRL evaluation strategies. We then validate the approach through two illustrative examples based on real-world projects. The results from the illustrative examples demonstrate that the data requirements identified by the proposed approach align with the requirements of real-world projects, demonstrating the practicality and effectiveness of the proposed framework. The proposed dataset selection customization mechanism and the proposed DRGM are helpful in guiding non-experts in identifying the data requirements for machine learning systems tailored to a specific ML problem. This approach also aids in evaluating different dataset alternatives to choose the optimum dataset for the problem. For future work, we recommend implementing tool support to generate the DRGM based on a chatbot interface.
Authors: Feijiang Han, Jiaming Zhang, Chuyi Deng, Jianheng Tang, Yunhuai Liu
Abstract: WebShell attacks, where malicious scripts are injected into web servers, pose a significant cybersecurity threat. Traditional ML and DL methods are often hampered by challenges such as the need for extensive training data, catastrophic forgetting, and poor generalization. Recently, Large Language Models have emerged as powerful alternatives for code-related tasks, but their potential in WebShell detection remains underexplored. In this paper, we make two contributions: (1) a comprehensive evaluation of seven LLMs, including GPT-4, LLaMA 3.1 70B, and Qwen 2.5 variants, benchmarked against traditional sequence- and graph-based methods using a dataset of 26.59K PHP scripts, and (2) the Behavioral Function-Aware Detection (BFAD) framework, designed to address the specific challenges of applying LLMs to this domain. Our framework integrates three components: a Critical Function Filter that isolates malicious PHP function calls, a Context-Aware Code Extraction strategy that captures the most behaviorally indicative code segments, and Weighted Behavioral Function Profiling that enhances in-context learning by prioritizing the most relevant demonstrations based on discriminative function-level profiles. Our results show that, stemming from their distinct analytical strategies, larger LLMs achieve near-perfect precision but lower recall, while smaller models exhibit the opposite trade-off. However, all baseline models lag behind previous SOTA methods. With the application of BFAD, the performance of all LLMs improves significantly, yielding an average F1 score increase of 13.82%. Notably, larger models now outperform SOTA benchmarks, while smaller models such as Qwen-2.5-Coder-3B achieve performance competitive with traditional methods. This work is the first to explore the feasibility and limitations of LLMs for WebShell detection and provides solutions to address the challenges in this task.
Authors: Yuji Okamoto, Tomoya Takeuchi, Yusuke Sakemi
Abstract: Since the advent of the ``Neural Ordinary Differential Equation (Neural ODE)'' paper, learning ODEs with deep learning has been applied to system identification, time-series forecasting, and related areas. Exploiting the diffeomorphic nature of ODE solution maps, neural ODEs has also enabled their use in generative modeling. Despite the rich potential to incorporate various kinds of physical information, training Neural ODEs remains challenging in practice. This study demonstrates, through the simplest one-dimensional linear model, why training Neural ODEs is difficult. We then propose a new stabilization method and provide an analytical convergence analysis. The insights and techniques presented here serve as a concise tutorial for researchers beginning work on Neural ODEs.
Authors: Nicholas Pfaff, Hongkai Dai, Sergey Zakharov, Shun Iwase, Russ Tedrake
Abstract: Training robots in simulation requires diverse 3D scenes that reflect the specific challenges of downstream tasks. However, scenes that satisfy strict task requirements, such as high-clutter environments with plausible spatial arrangement, are rare and costly to curate manually. Instead, we generate large-scale scene data using procedural models that approximate realistic environments for robotic manipulation, and adapt it to task-specific goals. We do this by training a unified diffusion-based generative model that predicts which objects to place from a fixed asset library, along with their SE(3) poses. This model serves as a flexible scene prior that can be adapted using reinforcement learning-based post training, conditional generation, or inference-time search, steering generation toward downstream objectives even when they differ from the original data distribution. Our method enables goal-directed scene synthesis that respects physical feasibility and scales across scene types. We introduce a novel MCTS-based inference-time search strategy for diffusion models, enforce feasibility via projection and simulation, and release a dataset of over 44 million SE(3) scenes spanning five diverse environments. Website with videos, code, data, and model weights: https://steerable-scene-generation.github.io/
Authors: Yanhui Hong, Nan Wang, Zhiyi Xia, Haoyi Tao, Xi Fang, Yiming Li, Jiankun Wang, Peng Jin, Xiaochen Cai, Shengyu Li, Ziqi Chen, Zezhong Zhang, Guolin Ke, Linfeng Zhang
Abstract: This paper presents a systematic solution for the intelligent recognition and automatic analysis of microscopy images. We developed a data engine that generates high-quality annotated datasets through a combination of the collection of diverse microscopy images from experiments, synthetic data generation and a human-in-the-loop annotation process. To address the unique challenges of microscopy images, we propose a segmentation model capable of robustly detecting both small and large objects. The model effectively identifies and separates thousands of closely situated targets, even in cluttered visual environments. Furthermore, our solution supports the precise automatic recognition of image scale bars, an essential feature in quantitative microscopic analysis. Building upon these components, we have constructed a comprehensive intelligent analysis platform and validated its effectiveness and practicality in real-world applications. This study not only advances automatic recognition in microscopy imaging but also ensures scalability and generalizability across multiple application domains, offering a powerful tool for automated microscopic analysis in interdisciplinary research. A online application is made available for researchers to access and evaluate the proposed automated analysis service.
Authors: Felipe Areces, John C. Duchi
Abstract: The basic question of delineating those statistical problems that are solvable without making any assumptions on the underlying data distribution has long animated statistics and learning theory. This paper characterizes when a convex M-estimation or stochastic optimization problem is solvable in such an assumption-free setting, providing a precise dividing line between solvable and unsolvable problems. The conditions we identify show, perhaps surprisingly, that Lipschitz continuity of the loss being minimized is not necessary for distribution free minimization, and they are also distinct from classical characterizations of learnability in machine learning.
Authors: Liyan Xu, Zhenlin Su, Mo Yu, Jiangnan Li, Fandong Meng, Jie Zhou
Abstract: This work stems from an observed limitation of text encoders: embeddings may not be able to recognize fine-grained entities or events within encoded semantics, resulting in failed retrieval even in simple cases. To examine such behaviors, we first introduce a new evaluation dataset, CapRetrieval, in which passages are image captions and queries are phrases targeting entity or event concepts in diverse forms. Zero-shot evaluation suggests that encoders often struggle with these fine-grained matching, regardless of training sources or model size. Aiming for enhancement, we proceed to finetune encoders with our proposed data generation strategies, enabling a small 0.1B encoder to outperform the state-of-the-art 7B model. Within this process, we further uncover the granularity dilemma, a challenge for embeddings to capture fine-grained salience while aligning with overall semantics. Our dataset, code and models in this work are publicly released at https://github.com/lxucs/CapRetrieval.
Authors: Shiyu Jiang, Xuyin Liu, Zitong Jerry Wang
Abstract: Genomic language models (gLMs) hold promise for generating novel, functional DNA sequences for synthetic biology. However, realizing this potential requires models to go beyond evolutionary plausibility and understand how DNA sequence encodes gene expression and regulation. We introduce a benchmark called Nullsettes, which assesses how well models can predict in silico loss-of-function (LOF) mutations, in synthetic expression cassettes with little evolutionary precedent. Testing 12 state-of-the-art gLMs, we find that most fail to consistently detect these strong LOF mutations. All models show a sharp drop in predictive accuracy as the likelihood assigned to the original (nonmutant) sequence decreases, suggesting that gLMs rely heavily on pattern-matching to their evolutionary prior rather than on any mechanistic understanding of gene expression. Our findings highlight fundamental limitations in how gLMs generalize to engineered, non-natural sequences, and underscore the need for benchmarks and modeling strategies that prioritize functional understanding.
Authors: Louise Largeau, Erwan Koch, David Leutwyler, Gregoire Mariethoz, Valerie Chavez-Demoulin, Tom Beucler
Abstract: The coarse spatial resolution of gridded climate models, such as general circulation models, limits their direct use in projecting socially relevant variables like extreme precipitation. Most downscaling methods estimate the conditional distributions of extremes by generating large ensembles, complicating the assessment of robustness under distributional shifts, such as those induced by climate change. To better understand and potentially improve robustness, we propose super-resolving the parameters of the target variable's probability distribution directly using analytically tractable mappings. Within a perfect-model framework over Switzerland, we demonstrate that vector generalized linear and additive models can super-resolve the generalized extreme value distribution of summer hourly precipitation extremes from coarse precipitation fields and topography. We introduce the notion of a "robustness gap", defined as the difference in predictive error between present-trained and future-trained models, and use it to diagnose how model structure affects the generalization of each quantile to a pseudo-global warming scenario. By evaluating multiple model configurations, we also identify an upper limit on the super-resolution factor based on the spatial auto- and cross-correlation of precipitation and elevation, beyond which coarse precipitation loses predictive value. Our framework is broadly applicable to variables governed by parametric distributions and offers a model-agnostic diagnostic for understanding when and why empirical downscaling generalizes to climate change and extremes.
Authors: Ammar Ahmed, Ali Shariq Imran, Zenun Kastrati, Sher Muhammad Daudpota
Abstract: Wrist pathologies are frequently observed, particularly among children who constitute the majority of fracture cases. Computer vision presents a promising avenue, contingent upon the availability of extensive datasets, a notable challenge in medical imaging. Therefore, reliance solely on one modality, such as images, proves inadequate, especially in an era of diverse and plentiful data types. In this study, we employ a multifaceted approach to address the challenge of recognizing wrist pathologies using an extremely limited dataset. Initially, we approach the problem as a fine-grained recognition task. Secondly, we enhance network performance by fusing patient metadata with X-rays. Thirdly, we improve the performance further by utilizing weights trained on a separate fine-grained dataset. While metadata integration has been used in other medical domains, this is a novel application for wrist pathologies.
Authors: Terry Yue Zhuo, Dingmin Wang, Hantian Ding, Varun Kumar, Zijian Wang
Abstract: Large Language Models (LLMs) have achieved remarkable success in software engineering tasks when trained with executable runtime environments, particularly in resolving GitHub issues. However, such runtime environments are often unavailable in other domains, especially cybersecurity, where challenge configurations and execution contexts are ephemeral or restricted. We present Cyber-Zero, the first runtime-free framework for synthesizing high-quality agent trajectories to train cybersecurity LLMs. Cyber-Zero leverages publicly available CTF writeups and employs persona-driven LLM simulation to reverse-engineer runtime behaviors and generate realistic, long-horizon interaction sequences without actual environments. Using trajectories synthesized by Cyber-Zero, we train LLM-based agents that achieve up to 13.1% absolute performance gains over baseline models on three prominent CTF benchmarks: InterCode-CTF, NYU CTF Bench, and Cybench. Our best model, Cyber-Zero-32B, establishes new state-of-the-art performance among open-weight models, matching the capabilities of proprietary systems like DeepSeek-V3-0324 and Claude-3.5-Sonnet while offering superior cost-effectiveness, and demonstrating that runtime-free trajectory synthesis can effectively democratize the development of state-of-the-art cybersecurity agents.
Authors: Jaewan Moon, Seongmin Park, Jongwuk Lee
Abstract: Large language models (LLMs) have been widely adopted to enrich the semantic representation of textual item information in recommender systems. However, existing linear autoencoders (LAEs) that incorporate textual information rely on sparse word co-occurrence patterns, limiting their ability to capture rich textual semantics. To address this, we propose L3AE, the first integration of LLMs into the LAE framework. L3AE effectively integrates the heterogeneous knowledge of textual semantics and user-item interactions through a two-phase optimization strategy. (i) L3AE first constructs a semantic item-to-item correlation matrix from LLM-derived item representations. (ii) It then learns an item-to-item weight matrix from collaborative signals while distilling semantic item correlations as regularization. Notably, each phase of L3AE is optimized through closed-form solutions, ensuring global optimality and computational efficiency. Extensive experiments demonstrate that L3AE consistently outperforms state-of-the-art LLM-enhanced models on three benchmark datasets, achieving gains of 27.6% in Recall@20 and 39.3% in NDCG@20. The source code is available at https://github.com/jaewan7599/L3AE_CIKM2025.
Authors: NVIDIA, :, Aarti Basant, Abhijit Khairnar, Abhijit Paithankar, Abhinav Khattar, Adithya Renduchintala, Aditya Malte, Akhiad Bercovich, Akshay Hazare, Alejandra Rico, Aleksander Ficek, Alex Kondratenko, Alex Shaposhnikov, Alexander Bukharin, Ali Taghibakhshi, Amelia Barton, Ameya Sunil Mahabaleshwarkar, Amy Shen, Andrew Tao, Ann Guan, Anna Shors, Anubhav Mandarwal, Arham Mehta, Arun Venkatesan, Ashton Sharabiani, Ashwath Aithal, Ashwin Poojary, Ayush Dattagupta, Balaram Buddharaju, Banghua Zhu, Barnaby Simkin, Bilal Kartal, Bita Darvish Rouhani, Bobby Chen, Boris Ginsburg, Brandon Norick, Brian Yu, Bryan Catanzaro, Charles Wang, Charlie Truong, Chetan Mungekar, Chintan Patel, Chris Alexiuk, Christian Munley, Christopher Parisien, Dan Su, Daniel Afrimi, Daniel Korzekwa, Daniel Rohrer, Daria Gitman, David Mosallanezhad, Deepak Narayanan, Dima Rekesh, Dina Yared, Dmytro Pykhtar, Dong Ahn, Duncan Riach, Eileen Long, Elliott Ning, Eric Chung, Erick Galinkin, Evelina Bakhturina, Gargi Prasad, Gerald Shen, Haifeng Qian, Haim Elisha, Harsh Sharma, Hayley Ross, Helen Ngo, Herman Sahota, Hexin Wang, Hoo Chang Shin, Hua Huang, Iain Cunningham, Igor Gitman, Ivan Moshkov, Jaehun Jung, Jan Kautz, Jane Polak Scowcroft, Jared Casper, Jian Zhang, Jiaqi Zeng, Jimmy Zhang, Jinze Xue, Jocelyn Huang, Joey Conway, John Kamalu, Jonathan Cohen, Joseph Jennings, Julien Veron Vialard, Junkeun Yi, Jupinder Parmar, Kari Briski, Katherine Cheung, Katherine Luna, Keith Wyss, Keshav Santhanam, Kezhi Kong, Krzysztof Pawelec, Kumar Anik, Kunlun Li, Kushan Ahmadian, Lawrence McAfee, Laya Sleiman, Leon Derczynski, Luis Vega, Maer Rodrigues de Melo, Makesh Narsimhan Sreedhar, Marcin Chochowski, Mark Cai, Markus Kliegl, Marta Stepniewska-Dziubinska, Matvei Novikov, Mehrzad Samadi, Meredith Price, Meriem Boubdir, Michael Boone, Michael Evans, Michal Bien, Michal Zawalski, Miguel Martinez, Mike Chrzanowski, Mohammad Shoeybi, Mostofa Patwary, Namit Dhameja, Nave Assaf, Negar Habibi, Nidhi Bhatia, Nikki Pope, Nima Tajbakhsh, Nirmal Kumar Juluru, Oleg Rybakov, Oleksii Hrinchuk, Oleksii Kuchaiev, Oluwatobi Olabiyi, Pablo Ribalta, Padmavathy Subramanian, Parth Chadha, Pavlo Molchanov, Peter Dykas, Peter Jin, Piotr Bialecki, Piotr Januszewski, Pradeep Thalasta, Prashant Gaikwad, Prasoon Varshney, Pritam Gundecha, Przemek Tredak, Rabeeh Karimi Mahabadi, Rajen Patel, Ran El-Yaniv, Ranjit Rajan, Ria Cheruvu, Rima Shahbazyan, Ritika Borkar, Ritu Gala, Roger Waleffe, Ruoxi Zhang, Russell J. Hewett, Ryan Prenger, Sahil Jain, Samuel Kriman, Sanjeev Satheesh, Saori Kaji, Sarah Yurick, Saurav Muralidharan, Sean Narenthiran, Seonmyeong Bak, Sepehr Sameni, Seungju Han, Shanmugam Ramasamy, Shaona Ghosh, Sharath Turuvekere Sreenivas, Shelby Thomas, Shizhe Diao, Shreya Gopal, Shrimai Prabhumoye, Shubham Toshniwal, Shuoyang Ding, Siddharth Singh, Siddhartha Jain, Somshubra Majumdar, Soumye Singhal, Stefania Alborghetti, Syeda Nahida Akter, Terry Kong, Tim Moon, Tomasz Hliwiak, Tomer Asida, Tony Wang, Tugrul Konuk, Twinkle Vashishth, Tyler Poon, Udi Karpas, Vahid Noroozi, Venkat Srinivasan, Vijay Korthikanti, Vikram Fugro, Vineeth Kalluru, Vitaly Kurin, Vitaly Lavrukhin, Wasi Uddin Ahmad, Wei Du, Wonmin Byeon, Ximing Lu, Xin Dong, Yashaswi Karnati, Yejin Choi, Yian Zhang, Ying Lin, Yonggan Fu, Yoshi Suhara, Zhen Dong, Zhiyu Li, Zhongbo Zhu, Zijia Chen
Abstract: We introduce Nemotron-Nano-9B-v2, a hybrid Mamba-Transformer language model designed to increase throughput for reasoning workloads while achieving state-of-the-art accuracy compared to similarly-sized models. Nemotron-Nano-9B-v2 builds on the Nemotron-H architecture, in which the majority of the self-attention layers in the common Transformer architecture are replaced with Mamba-2 layers, to achieve improved inference speed when generating the long thinking traces needed for reasoning. We create Nemotron-Nano-9B-v2 by first pre-training a 12-billion-parameter model (Nemotron-Nano-12B-v2-Base) on 20 trillion tokens using an FP8 training recipe. After aligning Nemotron-Nano-12B-v2-Base, we employ the Minitron strategy to compress and distill the model with the goal of enabling inference on up to 128k tokens on a single NVIDIA A10G GPU (22GiB of memory, bfloat16 precision). Compared to existing similarly-sized models (e.g., Qwen3-8B), we show that Nemotron-Nano-9B-v2 achieves on-par or better accuracy on reasoning benchmarks while achieving up to 6x higher inference throughput in reasoning settings like 8k input and 16k output tokens. We are releasing Nemotron-Nano-9B-v2, Nemotron-Nano12B-v2-Base, and Nemotron-Nano-9B-v2-Base checkpoints along with the majority of our pre- and post-training datasets on Hugging Face.
Authors: Amrapali Pednekar, Alvaro Garrido, Pieter Simoens, Yara Khaluf
Abstract: Drawing parallels between Deep Artificial Neural Networks (DNNs) and biological systems can aid in understanding complex biological mechanisms that are difficult to disentangle. Temporal processing, an extensively researched topic, is one such example that lacks a coherent understanding of its underlying mechanisms. In this study, we investigate temporal processing in a Deep Reinforcement Learning (DRL) agent performing an interval timing task and explore potential biological counterparts to its emergent behavior. The agent was successfully trained to perform a duration production task, which involved marking successive occurrences of a target interval while viewing a video sequence. Analysis of the agent's internal states revealed oscillatory neural activations, a ubiquitous pattern in biological systems. Interestingly, the agent's actions were predominantly influenced by neurons exhibiting these oscillations with high amplitudes and frequencies corresponding to the target interval. Parallels are drawn between the agent's time-keeping strategy and the Striatal Beat Frequency (SBF) model, a biologically plausible model of interval timing. Furthermore, the agent maintained its oscillatory representations and task performance when tested on different video sequences (including a blank video). Thus, once learned, the agent internalized its time-keeping mechanism and showed minimal reliance on its environment to perform the timing task. A hypothesis about the resemblance between this emergent behavior and certain aspects of the evolution of biological processes like circadian rhythms, has been discussed. This study aims to contribute to recent research efforts of utilizing DNNs to understand biological systems, with a particular emphasis on temporal processing.
Authors: Yixuan Sun, Romain Egele, Sri Hari Krishna Narayanan, Luke Van Roekel, Carmelo Gonzales, Steven Brus, Balu Nadiga, Sandeep Madireddy, Prasanna Balaprakash
Abstract: Accurate simulations of the oceans are crucial in understanding the Earth system. Despite their efficiency, simulations at lower resolutions must rely on various uncertain parameterizations to account for unresolved processes. However, model sensitivity to parameterizations is difficult to quantify, making it challenging to tune these parameterizations to reproduce observations. Deep learning surrogates have shown promise for efficient computation of the parametric sensitivities in the form of partial derivatives, but their reliability is difficult to evaluate without ground truth derivatives. In this work, we leverage large-scale hyperparameter search and ensemble learning to improve both forward predictions, autoregressive rollout, and backward adjoint sensitivity estimation. Particularly, the ensemble method provides epistemic uncertainty of function value predictions and their derivatives, providing improved reliability of the neural surrogates in decision making.
Authors: Somraj Gautam, Abhirama Subramanyam Penamakuri, Abhishek Bhandari, Gaurav Harit
Abstract: We introduce MMCRICBENCH-3K, a benchmark for Visual Question Answering (VQA) on cricket scorecards, designed to evaluate large vision-language models (LVLMs) on complex numerical and cross-lingual reasoning over semi-structured tabular images. MMCRICBENCH-3K comprises 1,463 synthetically generated scorecard images from ODI, T20, and Test formats, accompanied by 1,500 English QA pairs. It includes two subsets: MMCRICBENCH-E-1.5K, featuring English scorecards, and MMCRICBENCH-H-1.5K, containing visually similar Hindi scorecards, with all questions and answers kept in English to enable controlled cross-script evaluation. The task demands reasoning over structured numerical data, multi-image context, and implicit domain knowledge. Empirical results show that even state-of-the-art LVLMs, such as GPT-4o and Qwen2.5VL, struggle on the English subset despite it being their primary training language and exhibit a further drop in performance on the Hindi subset. This reveals key limitations in structure-aware visual text understanding, numerical reasoning, and cross-lingual generalization. The dataset is publicly available via Hugging Face at https://huggingface.co/datasets/DIALab/MMCricBench, to promote LVLM research in this direction.
Authors: Daniel Frees, Moritz Bolling, Aditri Bhagirath
Abstract: Modern computer vision models have proven to be highly useful for medical imaging classification and segmentation tasks, but the scarcity of medical imaging data often limits the efficacy of models trained from scratch. Transfer learning has emerged as a pivotal solution to this, enabling the fine-tuning of high-performance models on small data. Mei et al. (2022) found that pre-training CNNs on a large dataset of radiologist-labeled images (RadImageNet) enhanced model performance on downstream tasks compared to ImageNet pretraining. The present work extends Mei et al. (2022) by conducting a comprehensive investigation to determine optimal CNN architectures for breast lesion malignancy detection and ACL tear detection, as well as performing statistical analysis to compare the effect of RadImageNet and ImageNet pre-training on downstream model performance. Our findings suggest that 1-dimensional convolutional classifiers with skip connections, ResNet50 pre-trained backbones, and partial backbone unfreezing yields optimal downstream medical classification performance. Our best models achieve AUCs of 0.9969 for ACL tear detection and 0.9641 for breast nodule malignancy detection, competitive with the results reported by Mei et al. (2022) and surpassing other previous works. We do not find evidence confirming RadImageNet pre-training to provide superior downstream performance for ACL tear and breast lesion classification tasks.
Authors: Daniel Frees, Martin Pollack
Abstract: Despite the overwhelming performance improvements offered by recent natural language processing (NLP) models, the decisions made by these models are largely a black box. Towards closing this gap, the field of causal NLP combines causal inference literature with modern NLP models to elucidate causal effects of text features. We replicate and extend Bansal et al's work on regularizing text classifiers to adhere to estimated effects, focusing instead on model interpretability. Specifically, we focus on developing a two-headed RieszNet-based neural network architecture which achieves better treatment effect estimation accuracy. Our framework, CausalSent, accurately predicts treatment effects in semi-synthetic IMDB movie reviews, reducing MAE of effect estimates by 2-3x compared to Bansal et al's MAE on synthetic Civil Comments data. With an ensemble of validated models, we perform an observational case study on the causal effect of the word "love" in IMDB movie reviews, finding that the presence of the word "love" causes a +2.9% increase in the probability of a positive sentiment.
Authors: Bin Tan, Wangyao Ge, Yidi Wang, Xin Liu, Jeff Burtoft, Hao Fan, Hui Wang
Abstract: Modern app store recommender systems struggle with multiple-category apps, as traditional taxonomies fail to capture overlapping semantics, leading to suboptimal personalization. We propose PCR-CA (Parallel Codebook Representations with Contrastive Alignment), an end-to-end framework for improved CTR prediction. PCR-CA first extracts compact multimodal embeddings from app text, then introduces a Parallel Codebook VQ-AE module that learns discrete semantic representations across multiple codebooks in parallel -- unlike hierarchical residual quantization (RQ-VAE). This design enables independent encoding of diverse aspects (e.g., gameplay, art style), better modeling multiple-category semantics. To bridge semantic and collaborative signals, we employ a contrastive alignment loss at both the user and item levels, enhancing representation learning for long-tail items. Additionally, a dual-attention fusion mechanism combines ID-based and semantic features to capture user interests, especially for long-tail apps. Experiments on a large-scale dataset show PCR-CA achieves a +0.76% AUC improvement over strong baselines, with +2.15% AUC gains for long-tail apps. Online A/B testing further validates our approach, showing a +10.52% lift in CTR and a +16.30% improvement in CVR, demonstrating PCR-CA's effectiveness in real-world deployment. The new framework has now been fully deployed on the Microsoft Store.
Authors: Le Zhang, Fuping Wu, Arun Thirunavukarasu, Kevin Bronik, Thomas Nichols, Bartlomiej W. Papiez
Abstract: Large annotated datasets are vital for training segmentation models, but pixel-level labeling is time-consuming, error-prone, and often requires scarce expert annotators, especially in medical imaging. In contrast, coarse annotations are quicker, cheaper, and easier to produce, even by non-experts. In this paper, we propose to use coarse drawings from both positive (target) and negative (background) classes in the image, even with noisy pixels, to train a convolutional neural network (CNN) for semantic segmentation. We present a method for learning the true segmentation label distributions from purely noisy coarse annotations using two coupled CNNs. The separation of the two CNNs is achieved by high fidelity with the characters of the noisy training annotations. We propose to add a complementary label learning that encourages estimating negative label distribution. To illustrate the properties of our method, we first use a toy segmentation dataset based on MNIST. We then present the quantitative results of experiments using publicly available datasets: Cityscapes dataset for multi-class segmentation, and retinal images for medical applications. In all experiments, our method outperforms state-of-the-art methods, particularly in the cases where the ratio of coarse annotations is small compared to the given dense annotations.