Scholar
Michael Dennis
Google Scholar ID: WXXu26AAAAAJ
Google DeepMind
Citations & Impact
Citations
2,073
H-index
16
i10-index
18
Publications
20
Co-authors
0
Publications
Resume
Academic Achievements
- Published several papers, including:
- - PAIRED: Presented at NeurIPS 2020 (top 1% of submissions), which introduces a method to find minimax regret policies through training an adversary to generate levels that are hard for the protagonist but easy for the antagonist.
- - Adversarial Policies: Investigated how deep reinforcement learning agents can be affected by adversarial strategies from other agents, demonstrating the existence of such policies in zero-sum games involving simulated humanoid robots.
Research Experience
- Currently a Research Scientist on Google Deepmind's Openendedness team. Previously, conducted research as a Ph.D. student at CHAI.
Education
- Ph.D. student at the Center for Human-Compatible AI (CHAI), advised by Stuart Russell. Prior research focused on computer science theory and computational geometry.
Background
- Interested in the intersection between problem specification and open-ended complexity, focusing on Unsupervised Environment Design (UED) to automatically build complex and challenging environments for promoting efficient learning and transfer. Also deeply involved in decision theory.
Miscellany
- Connects via Email, Twitter, Google Scholar, and GitHub.