This paper establishes a large deviation principle for the generalization error of interpolating classifiers in the overparametrized regime.

This paper presents a memory-efficient training stack for Mixture-of-Experts (MoE) models, combining and specializing parallelism techniques for maximal efficiency.

This paper proposes an ensemble of two source separators, FTRNN and TF-Locoformer, trained with mixture invariant training (MixIT), and introduces mixture-constrained max pooling (MCM) to improve bird…

The paper proves a stochastic comparison for Gaussian maxima, resolving the Weak Simplex Conjecture and proving the Simplex Mean Width Conjecture.

GLAM-SLAM is a real-time, decoupled Gaussian-splatting SLAM system for large-scale outdoor scenes with a robust feature-based frontend and structured sparse mapping representation.

The paper introduces LLMSurgeon, a framework that estimates the domain-level data mixture of a Large Language Model (LLM) using only generated text, thereby providing a post-hoc method to audit the mo…

This book provides a compact, derivation-oriented mathematical primer that connects major families of generative AI models, showing their underlying structural relationships.

The paper introduces BACH, a multi-interest two-tower retrieval model that uses a per-user mixture over heads, mitigating collapse and producing a per-user weighting of interests.

This paper improves the foundations of neural likelihood approximation for Bayesian inverse problems by making the learning problem strictly convex and showing convergence to the true likelihood.

This paper establishes a direct connection between Approximate Message Passing (AMP) and the Convex Gaussian Min-max Theorem (CGMT) for regularized linear regression and M-estimation.

The paper shows that the scaled overshoot of a minimum value in a Gaussian process converges to an exponential random variable as the minimum value increases, and identifies the limiting measure as an…

This paper introduces Distributional Random Forests, which replace mean-based CART splitting with criteria that compare full conditional response distributions in candidate children. The authors syste…

This paper proposes a learning-based method for sound field estimation using a physics-constrained neural kernel with a source-position-dependent INR for directional weighting function.

This paper analyzes the Bayesian fixed-budget best-arm identification problem with abstention, showing that it induces a phase transition from polynomial to exponential decay of error probability.

This paper develops scalable methods for time-series analysis using tensor algebra in factorial hidden Markov models, improving computational performance and enabling analysis of large systems.

A multimodal Mixture-of-Experts framework was developed for asthma detection using vocal biomarkers and clinical data, achieving better performance than unimodal and bimodal approaches.

This paper investigates the correlation between dimensions of self-supervised speech features and speech characteristics, finding that content dimensions primarily capture intensity, formants, and voi…

The paper introduces 'mixture mechanisms,' a novel class of additive noise mechanisms that achieve approximate differential privacy by mixing multiple Gaussian distributions, resulting in lower noise…

The paper introduces 'mixture mechanisms,' a novel class of additive noise mechanisms that achieve differential privacy for real-valued queries, significantly reducing noise compared to the standard G…

This paper investigates the impact of batch sampling strategies using text and audio embeddings on text-to-music generation under low-data conditions.