AI Theory

GNAN: a novel interpretable graph neural network achieving accuracy comparable to black-box models.

Leveraging data heterogeneity, this study reveals that standard SGD implicitly learns invariant features across multiple environments, achieving robust generalization without explicit regularization.

Deep learning models’ success hinges on understanding gradient descent’s implicit bias. This study reveals how this bias influences layer collaboration, revealing a decreasing trend in adversarial rob…

Adam’s implicit bias revealed: On separable data, Adam converges towards the maximum l∞-margin solution, a finding contrasting with gradient descent’s l2-margin preference. This polynomial-time conver…

Finetuning’s impact on worst-group accuracy is surprisingly nuanced, with common class-balancing methods sometimes hurting performance; a novel mixture method consistently outperforms others.

Novel algorithms trace the optimal balance between clustering quality and fairness, revealing all non-dominated solutions for various objectives.

Breaking neural network parameter symmetries leads to faster training, better generalization, and improved loss landscape behavior, as demonstrated by novel asymmetric network architectures.

Physics-Informed Neural Networks (PINNs) training dynamics for nonlinear PDEs are fundamentally different than linear ones; this paper reveals why using second-order methods is crucial for solving non…

Diverse inhibitory neurons in brain circuits enable faster Bayesian computation via Hamiltonian sampling.

This paper presents nearly-linear time algorithms for testing model calibration, improving upon existing methods and providing theoretical lower bounds for various calibration measures.

Testably learning polynomial threshold functions efficiently, matching agnostic learning’s best guarantees, is achieved, solving a key problem in robust machine learning.

Temp-G³NTK: a novel temporal graph neural tangent kernel guarantees convergence to graphon NTK, offering superior performance in temporal graph classification and node-level tasks.

Adaptive experiment design optimizes indirect experiments in complex systems by sequentially narrowing the gap between upper and lower bounds on a targeted query, providing more efficient and informat…

Tangent Space Causal Inference (TSCI) enhances causal discovery in dynamical systems by leveraging vector fields, outperforming existing methods in accuracy and interpretability.

Overparametrized neural networks’ learning dynamics are analyzed under data symmetries using mean-field theory, revealing that data augmentation, feature averaging, and equivariant architectures asymp…

SymILO: A novel symmetry-aware learning framework dramatically improves integer linear program (ILP) solutions by addressing data variability caused by ILP symmetry.

SureMap, a new method, significantly boosts accuracy in single and multi-task disaggregated evaluations of AI models using limited data by transforming the problem into Gaussian mean estimation and cl…

SuperDeepFool: a fast, accurate algorithm generating minimal adversarial perturbations, significantly improving deep learning model robustness evaluation and adversarial training.

Predicting outcomes of combined sequential interventions is challenging, especially in sparse data. This paper introduces CSI-VAE, a novel compositional model that provides reliable predictions for u…

Neuromodulated RNNs (NM-RNNs) enhance RNN flexibility by dynamically scaling recurrent weights using a neuromodulatory subnetwork, achieving higher accuracy and generalizability on various tasks compa…