Tom Zahavy

About

Tom Zahavy is a researcher specializing in **deep reinforcement learning (DRL)**, with a particular focus on developing analytical and visualization tools to better understand the behavior of DRL agents. At a time when reinforcement learning was demonstrating remarkable promise — achieving breakthroughs in Atari game playing, the game of Go, and robotic control — Zahavy recognized a critical gap: while these agents performed impressively in practice, the research community lacked robust tools to interpret and analyze their decision-making processes. His 2016 work, "Visualizing Dynamics: from t-SNE to SEMI-MDPs," addressed this challenge by introducing novel visualization techniques to illuminate the internal dynamics of DRL agents, earning 12 citations. Complementing this, his paper "Deep Reinforcement Learning Discovers Internal Models" further explored how DRL agents develop internal representations of their environments, contributing 3 additional citations to his growing body of work. Zahavy's contributions are especially valuable for students and practitioners seeking interpretability in reinforcement learning systems — an area that remains vitally important as AI agents are deployed in increasingly complex real-world scenarios. His work bridges the gap between raw agent performance and human-understandable insight into how these systems truly operate.

Research Focus

Key Achievements

Top Papers

1. 1
   Visualizing Dynamics: from t-SNE to SEMI-MDPs

   12 citations · 2016
2. 2
   Deep Reinforcement Learning Discovers Internal Models

   3 citations · 2016

Key Collaborators

Contact & Links