2026
High-Dimensional Interpolators Can Be Fragile: Heavy Tails and High-Dimensional Large Deviations
High dimensional large deviations and rare severe prediction risk.
03 · Publications
Selected papers and preprints.
Research across generative inference, scientific machine learning, optimization, probability, and computational mathematics.
ML Theory
2026
An AI-Assisted Solution to the Signed BAR Conjecture: Uniqueness in the Harrison--Reiman Class and a Completely-S Class Obstruction
Applied probability and AI assisted mathematical discovery.
ICML 2026
SURGE: Approximation and Training Free Particle Filter for Diffusion Surrogate
Particle filtering for diffusion surrogate inference.
ICML 2026
Simple Approximation and Derivative Free Inference-Time Scaling for Diffusion Models via Sequential Monte Carlo on Path Measures
Sequential Monte Carlo on diffusion path measures.
ICML 2026
RAGEN-2: Reasoning Collapse in Agentic RL
Understanding reasoning collapse in agentic reinforcement learning.
2026
When PINNs Go Wrong: Pseudo-Time Stepping Against Spurious Solutions
Pseudo time stepping as a safeguard against spurious solutions.
2026
On the Width Scaling of Neural Optimizers Under Matrix Operator Norms I: Row/Column Normalization and Hyperparameter Transfer
Matrix operator geometry and width stable optimization.
AISTATS 2026
A Covering Framework for Offline POMDPs Learning using Belief Space Metric
Belief space covering for offline learning under partial observability.
ICML 2026
On the Power of (Approximate) Reward Models for Inference-Time Scaling
Complexity theory for reward guided LLM reasoning.
ICLR 2026
Physics-Informed Inference Time Scaling for Solving High-Dimensional PDE via Defect Correction
Simulation calibrated correction for scientific machine learning.
AISTATS 2025
Randomized Iterative Solver as Iterative Refinement: A Simple Fix Towards Backward Stability
Backward stability for randomized linear solvers.
ICML 2024
Orthogonal Bootstrap: Efficient Simulation of Input Uncertainty
Efficient simulation and uncertainty quantification.
ICLR 2023
Minimax Optimal Kernel Operator Learning via Multilevel Training
Optimal multilevel methods for operator learning.
ICLR 2022
An Unconstrained Layer-Peeled Perspective on Neural Collapse
Optimization geometry behind neural collapse.
ICML 2018
PDE-Net: Learning PDEs from Data
Learning differential operators and dynamics from observations.
No publications match this search.