Spotlight AI Theories

Online fair division algorithm achieves Õ(T²/³) regret while guaranteeing envy-freeness or proportionality in expectation, a result proven tight.

Unbalanced initializations dramatically accelerate neural network feature learning by modifying the geometry of learning trajectories, enabling faster feature extraction and improved generalization.

Researchers derived tighter generalization bounds for label-specific representation learning (LSRL) methods, improving understanding of LSRL’s success and offering guidance for future algorithm develo…

VIJI, a novel second-order algorithm, achieves optimal convergence rates for variational inequalities even with inexact Jacobian information, bridging the gap between theory and practice in machine le…

MoE-BiEntIRL: A novel explainable inverse reinforcement learning method enhances GNN robustness against diverse social media attacks by reconstructing attacker policies and generating more robust trai…

GEBM: a novel graph-based energy model for robust GNN uncertainty estimation.

Generative models learn hidden capabilities suddenly during training, which can be explained and predicted using a novel ‘concept space’ framework that analyzes learning dynamics and concept signal.

ECLipsE: A novel compositional approach drastically accelerates Lipschitz constant estimation for deep neural networks, achieving speedups of thousands of times compared to the state-of-the-art while …

This work delivers dimension-free deterministic equivalents for random feature regression, revealing sharp excess error rates and scaling laws.

Neural networks, viewed as Bayesian ensembles of fixed classifiers, enable continual learning without forgetting; posterior updates mirror stochastic gradient descent, offering insights into optimizat…

Constrained Adaptive Attack (CAA) significantly improves adversarial attacks on deep learning models for tabular data by combining gradient and search-based methods, achieving up to 96.1% accuracy dro…

Pre-trained transformer models can predict human smell perception by encoding odorant chemical structures, aligning with expert labels, continuous ratings, and similarity assessments.

Leaky ReLU networks exhibit benign overfitting under surprisingly relaxed conditions: input dimension only needs to linearly scale with sample size, challenging prior assumptions in the field.

Randomness-efficient algorithms are developed for online decision making, requiring only 2log n random bits and achieving near-optimal competitiveness for metrical task systems.

This paper revolutionizes AI alignment by applying social choice theory axioms to RLHF, exposing flaws in existing methods and proposing novel, axiomatically guaranteed reward learning rules.

MC-EIF automates efficient statistical estimation for high-dimensional models, integrating seamlessly with existing differentiable probabilistic programming systems and achieving optimal convergence r…

New metrics audit label privatization, revealing differentially private schemes often outperform heuristic methods in the privacy-utility tradeoff.

Graph Neural Networks (GNNs) learn optimal approximation algorithms for combinatorial optimization problems, achieving high-quality solutions for Max-Cut, Min-Vertex-Cover, and Max-3-SAT, while also p…

Random-order stream data necessitates efficient top eigenvector approximation; this paper presents novel algorithms with improved space complexity, achieving near-optimal bounds.

Latent Mutual Information (LMI) approximation accurately estimates mutual information in high-dimensional data using low-dimensional learned representations, solving a critical problem in various scie…