Publications

First-author ICLR 2026 paper introducing GraphBSI, a one-shot graph generative model based on Bayesian Sample Inference. Achieves state-of-the-art molecular and synthetic graph generation on the Moses and GuacaMol benchmarks, with ~12 GPU-months of experiments in PyTorch.

Co-authored work that uses deep reinforcement learning (PPO) to compute and verify equilibria in finite-horizon dynamic oligopoly games, showing predatory pricing can emerge as a rational equilibrium strategy under asymmetric costs.