10589 papers
This paper analyzes the convergence and escape dynamics of Stochastic Gradient Descent in one-dimensional landscapes, establishing that while SGD reliably converges to local minima, it may linger near local maxima depending on noise variance and geometry, with specific results provided for the probability of escaping sharp maxima to neighboring minima.
This paper introduces the BAH dataset, a multimodal collection of 1,427 videos from 300 participants annotated for ambivalence and hesitancy recognition, alongside baseline benchmarking results that highlight the need for advanced models to support personalized digital health interventions.
SafeDPO is a lightweight, theory-driven method that achieves provably optimal safety alignment in Large Language Models by deriving a closed-form solution for safety-constrained objectives, thereby eliminating the need for complex reward models or multi-stage pipelines while maintaining competitive helpfulness.
This paper introduces TADA, a targeted diffusion-based augmentation framework that selectively generates synthetic images for hard-to-learn examples to improve classifier generalization with significantly reduced computational overhead compared to full-dataset augmentation.
This paper proposes a computationally efficient supervised filter based on a Gumbel-copula implied upper-tail concordance score to identify features that are simultaneously extreme with the positive class, demonstrating its effectiveness in ranking clinically relevant predictors for diabetes risk across large-scale and clinical datasets while outperforming standard filters and matching strong baselines.
This paper introduces Supervised Calibration (SC), a loss-minimization framework that enhances In-Context Learning in Large Language Models by learning optimal per-class affine transformations to correct systematic biases and alter decision boundary orientations, thereby achieving state-of-the-art performance across multiple models and datasets.
This paper reviews the historical disconnect between approximation theory and machine learning practice, discusses emerging trends like deep networks and transformers, and introduces novel research enabling function approximation on unknown manifolds without requiring explicit manifold feature learning.
This paper presents a low-power, cost-effective diffractive optical system that integrates a passive diffractive layer with a shallow neural network to remotely and accurately reconstruct 3D structural vibration spectra, overcoming the scalability and complexity limitations of traditional Structural Health Monitoring solutions.
The paper presents AutoQD, a theoretically grounded method that automatically discovers diverse, high-performing policies in continuous control tasks by generating behavioral descriptors through random Fourier feature embeddings of policy occupancy measures, thereby eliminating the need for hand-crafted descriptors in Quality-Diversity optimization.
This paper introduces Conflict-Aware Evidential Deep Learning (C-EDL), a lightweight post-hoc method that enhances the robustness of uncertainty quantification against adversarial and out-of-distribution inputs by leveraging diverse task-preserving transformations to detect representational conflict and calibrate predictions without retraining.
This paper challenges the default use of honest estimation in causal forests, demonstrating through extensive benchmarking that while it prevents overfitting, it often increases underfitting and reduces the accuracy of individual-level treatment effect estimates, suggesting its application should be guided by specific goals and empirical evaluation rather than reflexive adoption.
This paper proposes a novel Bayesian framework that generalizes federated ADMM by leveraging variational inference duality, yielding both a theoretical unification of ADMM with Gaussian assumptions and practical, high-performance variants like Newton-like and Adam-like updates for diverse distribution families.
This paper presents a theoretical framework demonstrating that standard sparse autoencoders generally fail to recover ground truth monosemantic features from superposed polysemantic ones, and proposes a reweighted variant (WSAE) with a derived selection principle that significantly improves feature recovery and interpretability.
This paper proposes a novel context biasing method for automatic speech recognition that leverages user-provided on-the-fly corrections of substitution errors to effectively resolve pronunciation-orthography mismatches, achieving a 22% to 34% relative improvement in biased word error rates without compromising overall system performance.
Meta FAIR introduces UMA, a family of universal models for atoms trained on half a billion 3D structures using a novel mixture of linear experts architecture, which achieves state-of-the-art speed and accuracy across diverse chemical domains without requiring fine-tuning.
The paper introduces UQLM, a Python package that leverages state-of-the-art uncertainty quantification techniques to generate confidence scores for detecting hallucinations and enhancing the reliability of Large Language Model outputs.
This paper proposes Q-SVMPC, a novel framework that integrates Q-guided Stein variational inference with an RL-informed policy prior to enable diverse, robust, and sample-efficient trajectory optimization in Model Predictive Control, overcoming the mode collapse limitations of existing learning-based MPC methods.
This paper introduces Fast Equivariant Imaging (FEI), a novel unsupervised learning framework that leverages the Augmented Lagrangian method and auxiliary Plug-and-Play denoisers to achieve a 10x training acceleration and improved generalization for deep imaging tasks like X-ray CT reconstruction and inpainting without requiring ground-truth data.
This paper introduces a probabilistic early-exit framework for single-channel speech separation and enhancement that dynamically scales computational resources based on uncertainty-aware signal quality estimates, enabling efficient deployment on heterogeneous devices without compromising reconstruction performance.
This paper employs interpretability techniques on the off-by-one addition task to reveal that large language models achieve task-level generalization through a reusable "function induction" mechanism, where multiple attention heads collaboratively learn and compose abstract functions to solve unseen problems.