🏢 University of Michigan

Researchers developed a causally-motivated approach for ranking agents based on their gaming propensity, addressing the challenge of identifying ‘worst offenders’ in strategic classification settings.

This paper introduces a novel method for evaluating weakly supervised models using Fréchet bounds, providing reliable performance bounds without ground truth labels.

Diffusion models’ surprising generalizability stems from an inductive bias towards learning Gaussian data structures, a finding that reshapes our understanding of their training and generalization.

Researchers extended implicit bias theory to multiclass classification using a novel framework, proving that gradient descent prefers simple solutions even with complex alternatives.

Smoothed online classification can be harder than batch classification when label spaces are unbounded, challenging existing assumptions in machine learning.

Enhance Transformer models via Selective Self-Attention (SSA), a principled context control method that boosts accuracy and efficiency.

RoME, a robust contextual bandit algorithm, leverages mixed-effects modeling and debiased machine learning to optimize personalized mobile health interventions, achieving superior performance in simul…

This paper pioneers a modular framework for reinforcement learning, addressing the challenge of learning under complex observations and simpler latent dynamics, offering both statistical and algorithm…

PreTS algorithm efficiently identifies the most preferred policy in bandit problems with vector-valued rewards, achieving asymptotically optimal sample complexity.

Online learning algorithms can now leverage predictions about future data to achieve significantly lower regret, smoothly transitioning between worst-case and best-case performance based on prediction…

Graph Neural Networks (GNNs) struggle with heterophilic link prediction; this paper introduces formal definitions, theoretical analysis, improved designs, and real-world benchmarks to address this cha…

This paper proposes a computationally efficient, differentially private best subset selection method for high-dimensional sparse linear regression, achieving both strong statistical utility and provab…

LVLMs often hallucinate objects, a problem worsened when multiple objects are present. This paper introduces ROPE, a novel automated evaluation protocol that reveals how object class distribution and…

PaDIS: Patch-based diffusion inverse solver learns efficient image priors from image patches, enabling high-resolution inverse problem solutions with reduced computational costs and data needs.

Learn-To-be-Efficient (LTE) trains LLMs to achieve structured sparsity, boosting inference speed by 25% at 50% sparsity without sacrificing accuracy.

Ignorance is bliss: A new framework shows ignoring label noise in multi-class classification can achieve state-of-the-art performance, especially when using self-supervised feature extraction.

Place3D optimizes LiDAR placement for superior 3D scene understanding.

Researchers create ‘images that sound’—visual spectrograms looking like natural images and sounding like natural audio—by cleverly composing pre-trained image and audio diffusion models in a shared la…

aSeqDIP: A new unsupervised image reconstruction method using sequential deep image priors, achieving competitive performance with fewer data needs and faster runtimes.

Researchers achieve globally convergent variational inference by minimizing the expected forward KL divergence, overcoming the limitations of traditional methods.