Hello, I am Kazuki Ota. At ICML 2026, I presented two papers: one on stable and efficient self-play reinforcement learning for two-player games at the main conference, and one on self-supervised theorem discovery from axioms alone at the AI for Math Workshop.
ICML 2026 Main Conference Paper
Jul 8, 2:30 PM-4:15 PM, Hall A #306
We revisit regularized policy optimization for two-player games and develop KLENT, a search-free self-play reinforcement learning method that improves training stability and efficiency.
project page | paper | poster | event page on web | event page on mobile
Self-Supervised Theorem Discovery in a Formal Axiomatic System
3rd AI for Math Workshop
Jul 11, 11:05 AM-12:05 PM, Hall D1
We study whether useful theorems can emerge from axioms alone, and show that extracted theorems can serve as reusable lemmas for both the agent's own proof search and LLM-based proof search.
I am especially interested in discussing how learning systems can search for, test, and organize new knowledge beyond direct human supervision.
I also made an unofficial one-page calendar for ICML 2026. Feel free to use it.