Research
(*) denotes equal contribution, listed alphabetically as is customary in theoretical computer science.
2025
- arXiv Synthetic Error Injection Fails to Elicit Self-Correction In Language Models arXiv arXiv
- arXiv Markov Chains Approximate Message Passing arXiv arXiv
- STOC Weak Poincaré Inequalities, Simulated Annealing, and Sampling from Spherical Spin Glasses ACM Symposium on Theory of Computing arXiv
- ICLR Provable Weak-to-Strong Generalization via Benign Overfitting International Conference on Learning Representations arXiv
2024
- FOCS Locally Stationary Distributions: A Framework for Analyzing Slow-Mixing Markov Chains IEEE Annual Symposium on Foundations of Computer Science arXiv
- FOCS Fast Mixing in Sparse Random Ising Models IEEE Annual Symposium on Foundations of Computer Science arXiv
- STOC Robust recovery for stochastic block models, simplified and generalized ACM Symposium on Theory of Computing arXiv
2023
- NeurIPS Precise Asymptotic Generalization for Multiclass Classification with Overparameterized Linear Models Neural Information Processing Systems arXiv
- ICML On the Training Instability of Shuffling SGD with Batch Normalization International Conference on Machine Learning arXiv
- ISIT Lower Bounds for Multiclass Classification with Overparameterized Linear Models International Symposium on Information Theory
2022
- SIOPT Maximum A Posteriori Inference of Random Dot Product Graphs via Conic Programming SIAM Journal on Optimization arXiv