AI Theory

Unlocking latent community structures from multiple correlated networks is now possible with greater precision, as this research pinpoints the information-theoretic threshold for exact recovery, even …

HardCore: Fast generation of hard, realistic UNSAT problems for improved SAT solver runtime prediction.

Oracle-efficient algorithms conquer online multi-group learning, achieving sublinear regret even with massive, overlapping groups, paving the way for fair and efficient large-scale online systems.

GREAT Score: A novel framework using generative models for efficiently and accurately evaluating the global robustness of machine learning models against adversarial attacks.

GRAPHTRAIL unveils the first end-to-end global GNN explainer, translating black-box GNN predictions into easily interpretable boolean formulas over subgraph concepts, achieving significant improvement…

Graphcodes efficiently summarize complex datasets’ topological properties using graph neural networks, enhancing machine learning accuracy.

Graph Neural Networks’ (GNNs) computational power precisely mirrors that of arithmetic circuits, as proven via a novel C-GNN model; this reveals fundamental limits to GNN scalability.

This paper presents a novel optimistic mirror descent algorithm achieving optimal gradient-variation regret under generalized smoothness, applicable across convex, strongly convex functions, and fast-…

Novel gradient-based algorithms achieve O(√T) regret and O(T3/4) constraint violation for online DR-submodular maximization with stochastic long-term constraints.

This paper provides a novel optimization framework for guided diffusion models, proving Õ(1/K) convergence for concave objective functions and demonstrating structure-preserving guidance.

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

Reward-driven distortions in grid cell patterns are global, not local, preserving path integration while encoding environmental landmarks in spatial navigation.

Large-scale Transformer training’s global convergence is proven using weight decay regularization and a refined mean-field analysis, bridging theory and practice.

GLinSAT: A novel neural network layer efficiently solves general linear constraint satisfaction problems via accelerated gradient descent, enabling differentiable backpropagation and improved GPU perf…

UMOD: a novel multi-objective optimization algorithm efficiently generates uniformly distributed Pareto-optimal solutions by maximizing minimal pairwise distances, providing high-quality representatio…

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

New, tighter generalization bounds are derived for a class of stochastic learning algorithms that generate absolutely continuous probability distributions; enhancing our understanding of their perform…

Two-stage recommender systems using tree structures achieve better generalization with more branches and harmonized training data distributions across stages.

New information-theoretic generalization bounds, based on conditional f-information, improve existing methods by addressing unboundedness and offering a generic approach applicable to various loss fun…

Learn to project, solve faster! This paper introduces data-driven projections for solving high-dimensional linear programs, proving theoretical guarantees and demonstrating significant improvements in…