AI Theory

This paper presents novel theoretical results enabling the analytical calculation of Partial Information Decomposition for various probability distributions, including those relevant to neuroscience, …

Deep learning model interpretability improved via Sparse Rate Reduction (SRR), showing improved generalization and offering principled model design.

Dynamic regret minimization equals static regret in an extended space; this equivalence reveals a trade-off between loss variance and comparator variability, leading to a new algorithm achieving impro…

New research quantifies how search engines steer web traffic by subtly changing results, offering a powerful method for antitrust investigations and digital market analysis.

New unbiased gradient estimator for high-dimensional SDEs drastically reduces computation time without sacrificing estimation accuracy.

This paper introduces a unified ITR estimation framework using covariate balancing weights, achieving significant gains in robustness and effectiveness compared to existing methods.

Elo rating system’s convergence rigorously analyzed via Markov chains under the Bradley-Terry-Luce model, demonstrating competitive learning rates and informing efficient tournament design.

Accelerate neural network training using ‘amortized eigendecomposition’ – a novel method replacing expensive eigendecomposition with faster QR decomposition while preserving accuracy.

Almost-linear RNNs (AL-RNNs) offer highly interpretable symbolic codes for dynamical systems reconstruction, simplifying the analysis of complex systems.

Many graph neural networks (GNNs) surprisingly converge to constant outputs with increasing graph size, limiting their expressiveness.

Generalized linear bandits with subexponential reward distributions are self-concordant, enabling second-order regret bounds free of exponential dependence on problem parameters.

Post-processing pre-trained models for alignment using conformal risk control and property testing guarantees better alignment, even when training data is biased.

This paper introduces Quantitative Relative Judgment Aggregation (QRJA), a novel social choice model, and applies it to ranking prediction, yielding effective and interpretable results on various real…

Improved adversarially robust streaming algorithms for L_p estimation are presented, surpassing previous state-of-the-art space bounds and disproving the existence of inherent barriers.

Adversarially Robust Decision Transformer (ARDT) enhances offline RL robustness against powerful adversaries by conditioning policies on minimax returns, achieving superior worst-case performance.

Researchers refine Pearson’s correlation coefficient to precisely measure arbitrary monotone dependence, expanding its applicability beyond linear relationships.

This paper introduces efficient algorithms for online selection with a budget constraint when comparing candidates from different groups has a cost, improving fairness and efficiency.

Adaptive gradient descent methods are improved by leveraging local curvature information for entirely adaptive algorithms without added computational cost, proving convergence with only local Lipschit…

Adaptive experimentation tackles confounding in online A/B tests using encouragement designs and a novel linear bandit approach, achieving near-optimal sample complexity.

New adaptive second-order optimistic methods for minimax optimization achieve optimal convergence without line search, simplifying updates and improving efficiency.