theoretical results
Findings derived from mathematical proofs and analyses that provide insights into the properties and performance of AI algorithms without relying solely on empirical data.
- Block-Biased Mamba for Long-Range Sequence Processing
- CoLT: The conditional localization test for assessing the accuracy of neural posterior estimates
- Diffusion Models and the Manifold Hypothesis: Log-Domain Smoothing is Geometry Adaptive
- Effects of Dropout on Performance in Long-range Graph Learning Tasks
- Fairness under Competition
- Homogeneous Algorithms Can Reduce Competition in Personalized Pricing
- Learning from Delayed Feedback in Games via Extra Prediction
- Majority of the Bests: Improving Best-of-N via Bootstrapping
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers
- New Perspectives on the Polyak Stepsize: Surrogate Functions and Negative Results
- Optimality and NP-Hardness of Transformers in Learning Markovian Dynamical Functions
- RGNMR: A Gauss-Newton method for robust matrix completion with theoretical guarantees
- Risk Bounds For Distributional Regression
- Self-Boost via Optimal Retraining: An Analysis via Approximate Message Passing
- Sharp Gaussian approximations for Decentralized Federated Learning
- Sketched Gaussian Mechanism for Private Federated Learning
- Split conformal classification with unsupervised calibration
- Towards Generalizable Detector for Generated Image