Tabular data dominates data science but poses challenges for generative models, especially when the data is limited or sensitive. We present a novel approach to generating synthetic tabular data based on the principle of maximum entropy -- MaxEnt -- called GEM-T, for ``generative entropy maximization for tables.'' GEM-T directly captures nth-order interactions -- pairwise, third-order, etc. -- among columns of training data. In extensive testing, GEM-T matches or exceeds deep neural network approaches previously regarded as state-of-the-art in 23 of 34 publicly available datasets representing diverse subject domains (68\%). Notably, GEM-T involves orders-of-magnitude fewer trainable parameters, demonstrating that much of the information in real-world data resides in low-dimensional, potentially human-interpretable correlations, provided that the input data is appropriately transformed first. Furthermore, MaxEnt better handles heterogeneous data types (continuous vs. discrete vs. categorical), lack of local structure, and other features of tabular data. GEM-T represents a promising direction for light-weight high-performance generative models for structured data.

Hamiltonian dynamics describe a wide range of physical systems. As such, data-driven simulations of Hamiltonian systems are important for many scientific and engineering problems. In this work, we propose kernel-based methods for identifying and forecasting Hamiltonian systems directly from data. We present two approaches: a two-step method that reconstructs trajectories before learning the Hamiltonian, and a one-step method that jointly infers both. Across several benchmark systems, including mass-spring dynamics, a nonlinear pendulum, and the Henon-Heiles system, we demonstrate that our framework achieves accurate, data-efficient predictions and outperforms two-step kernel-based baselines, particularly in scarce-data regimes, while preserving the conservation properties of Hamiltonian dynamics. Moreover, our methodology provides theoretical a priori error estimates, ensuring reliability of the learned models. We also provide a more general, problem-agnostic numerical framework that goes beyond Hamiltonian systems and can be used for data-driven learning of arbitrary dynamical systems.

Generative models for counterfactual outcomes face two key sources of bias. Confounding bias arises when approaches fail to account for systematic differences between those who receive the intervention and those who do not. Misspecification bias arises when methods attempt to address confounding through estimation of an auxiliary model, but specify it incorrectly. We introduce DoubleGen, a doubly robust framework that modifies generative modeling training objectives to mitigate these biases. The new objectives rely on two auxiliaries -- a propensity and outcome model -- and successfully address confounding bias even if only one of them is correct. We provide finite-sample guarantees for this robustness property. We further establish conditions under which DoubleGen achieves oracle optimality -- matching the convergence rates standard approaches would enjoy if interventional data were available -- and minimax rate optimality. We illustrate DoubleGen with three examples: diffusion models, flow matching, and autoregressive language models.

A B S T R A C T We study the Evolutionary Deep Neural Network (EDNN) framework for accelerating numerical solvers of time-dependent partial differential equations (PDEs). We introduce a Low-Rank Evolutionary Deep Neural Network (LR-EDNN), which constrains parameter evolution to a low-rank subspace, thereby reducing the effective dimensionality of training while preserving solution accuracy. The low-rank tangent subspace is defined layer-wise by the singular value decomposition (SVD) of the current network weights, and the resulting update is obtained by solving a well-posed, tractable linear system within this subspace. We evaluate LR-EDNN on representative PDE problems and compare it against corresponding baselines. Across cases, LR-EDNN achieves comparable accuracy with substantially fewer trainable parameters and reduced computational cost. These results indicate that low-rank constraints on parameter velocities, rather than full-space updates, provide a practical path toward scalable, efficient, and reproducible scientific machine learning for PDEs. Introduction The application of deep learning to solving partial differential equations (PDEs) has emerged as an active and promising research area, providing a powerful alternative to traditional numerical methods. Unlike classical approaches such as finite difference, finite element, or spectral methods, which rely on discretization and iterative solvers, deep learning methods leverage neural networks to approximate solutions directly, often bypassing the need for meshes and offering flexibility in handling irregular domains and high-dimensional problems [1, 2]. Early efforts in this domain focused primarily on two paradigms.

Proper segmentation of organs-at-risk is important for radiation therapy, surgical planning, and diagnostic decision-making in medical image analysis. While deep learning-based segmentation architectures have made significant progress, they often fail to balance segmentation accuracy with computational efficiency. Most of the current state-of-the-art methods either prioritize performance at the cost of high computational complexity or compromise accuracy for efficiency. This paper addresses this gap by introducing an efficient dual-line decoder segmentation network (EDLDNet). The proposed method features a noisy decoder, which learns to incorporate structured perturbation at training time for better model robustness, yet at inference time only the noise-free decoder is executed, leading to lower computational cost. Multi-Scale convolutional Attention Modules (MSCAMs), Attention Gates (AGs), and Up-Convolution Blocks (UCBs) are further utilized to optimize feature representation and boost segmentation performance. By leveraging multi-scale segmentation masks from both decoders, we also utilize a mutation-based loss function to enhance the model's generalization. Our approach outperforms SOTA segmentation architectures on four publicly available medical imaging datasets. EDLDNet achieves SOTA performance with an 84.00% Dice score on the Synapse dataset, surpassing baseline model like UNet by 13.89% in Dice score while significantly reducing Multiply-Accumulate Operations (MACs) by 89.7%. Compared to recent approaches like EMCAD, our EDLDNet not only achieves higher Dice score but also maintains comparable computational efficiency. The outstanding performance across diverse datasets establishes EDLDNet's strong generalization, computational efficiency, and robustness. The source code, pre-processed data, and pre-trained weights will be available at https://github.com/riadhassan/EDLDNet .

Cross-attention is a core mechanism in encoder-decoder architectures, widespread in many fields, including speech-to-text (S2T) processing. Its scores have been repurposed for various downstream applications--such as timestamp estimation and audio-text alignment--under the assumption that they reflect the dependencies between input speech representation and the generated text. While the explanatory nature of attention mechanisms has been widely debated in the broader NLP literature, this assumption remains largely unexplored within the speech domain. To address this gap, we assess the explanatory power of cross-attention in S2T models by comparing its scores to input saliency maps derived from feature attribution. Our analysis spans monolingual and multilingual, single-task and multi-task models at multiple scales, and shows that attention scores moderately to strongly align with saliency-based explanations, particularly when aggregated across heads and layers. However, it also shows that cross-attention captures only about 50% of the input relevance and, in the best case, only partially reflects how the decoder attends to the encoder's representations--accounting for just 52-75% of the saliency. These findings uncover fundamental limitations in interpreting cross-attention as an explanatory proxy, suggesting that it offers an informative yet incomplete view of the factors driving predictions in S2T models.

End-to-end autonomous driving has substantially progressed by directly predicting future trajectories from raw perception inputs, which bypasses traditional modular pipelines. However, mainstream methods trained via imitation learning suffer from critical safety limitations, as they fail to distinguish between trajectories that appear human-like but are potentially unsafe. Some recent approaches attempt to address this by regressing multiple rule-driven scores but decoupling supervision from policy optimization, resulting in suboptimal performance. To tackle these challenges, we propose DriveDPO, a Safety Direct Preference Optimization Policy Learning framework. First, we distill a unified policy distribution from human imitation similarity and rule-based safety scores for direct policy optimization. Further, we introduce an iterative Direct Preference Optimization stage formulated as trajectory-level preference alignment. Extensive experiments on the NAVSIM benchmark demonstrate that DriveDPO achieves a new state-of-the-art PDMS of 90.0. Furthermore, qualitative results across diverse challenging scenarios highlight DriveDPO's ability to produce safer and more reliable driving behaviors.

Machine learning (ML) is rapidly transforming the way molecular dynamics simulations are performed and analyzed, from materials modeling to studies of protein folding and function. ML algorithms are often employed to learn low-dimensional representations of conformational landscapes and to cluster trajectories into relevant metastable states. Most of these algorithms require selecting a small number of features that describe the problem of interest. Although deep neural networks can tackle large numbers of input features, the training costs increase with input size, which makes the selection of a subset of features mandatory for most problems of practical interest. Here, we show that random nonlinear projections can be used to compress large feature spaces and make computations faster without substantial loss of information. We describe an efficient way to produce random projections and then exemplify the general procedure for protein folding. For our test cases NTL9 and the double-norleucin variant of the villin headpiece, we find that random compression retains the core static and dynamic information of the original high dimensional feature space and makes trajectory analysis more robust.

Large language models often generate factually incorrect outputs, motivating efforts to detect the truthfulness of their content. Most existing approaches rely on training probes over internal activations, but these methods suffer from scalability and generalization issues. A recent training-free method, NoVo, addresses this challenge by exploiting statistical patterns from the model itself. However, it focuses exclusively on attention mechanisms, potentially overlooking the MLP module-a core component of Transformer models known to support factual recall. In this paper, we show that certain value vectors within MLP modules exhibit truthfulness-related statistical patterns. Building on this insight, we propose TruthV, a simple and interpretable training-free method that detects content truthfulness by leveraging these value vectors. On the NoVo benchmark, TruthV significantly outperforms both NoVo and log-likelihood baselines, demonstrating that MLP modules-despite being neglected in prior training-free efforts-encode rich and useful signals for truthfulness detection. These findings offer new insights into how truthfulness is internally represented in LLMs and motivate further research on scalable and interpretable truthfulness detection.

Abstract--Current deep learning models for electroencephalog-raphy (EEG) are often task-specific and depend on large labeled datasets, limiting their adaptability. Although emerging foundation models aim for broader applicability, their rigid dependence on fixed, high-density multi-channel montages restricts their use across heterogeneous datasets and in missing-channel or practical low-channel settings. T o address these limitations, we introduce SingLEM, a self-supervised foundation model that learns robust, general-purpose representations from single-channel EEG, making it inherently hardware agnostic. The model employs a hybrid encoder architecture that combines convolutional layers to extract local features with a hierarchical transformer to model both short-and long-range temporal dependencies. When evaluated as a fixed feature extractor across six motor imagery and cognitive tasks, aggregated single-channel representations consistently outperformed leading multi-channel foundation models and handcrafted baselines. These results demonstrate that a single-channel approach can achieve state-of-the-art generalization while enabling fine-grained neurophysiological analysis and enhancing interpretability. LECTROENCEPHALOGRAPHY (EEG) is a noninvasive neurophysiological technique that measures brain activity through scalp electrodes. Because of to its high temporal resolution, portability, and affordability, EEG is widely applied in diverse domains, including brain-computer interfaces (BCIs) [1], sleep staging [2], seizure detection [3], [4], [5], clinical diagnosis [6], [7], and emotion recognition [8], [9], [10]. Despite its potential, EEG analysis is challenged by non-stationarity across subjects and sessions, susceptibility to noise (e.g., ocular or muscular artifacts), and low signal-to-noise ratios [11]. To address this, deep neural networks (DNNs) have emerged as the state-of-the-art paradigm, learning complex and task-relevant features automatically from raw data [12]. This work was supported in part by JSPS KAKENHI 23H00548. The work of Jamiyan Sukhbaatar was supported by the Mongolia-Japan Engineering for Education Development (MJEED) project.