Benjamin Van Roy

Timeline

1. 1993–1998

   Computer science degrees at MIT

   He earns an SB in Computer Science and Engineering at MIT in 1993 and an SM in Electrical Engineering and Computer Science in 1995. In 1998, he completes a PhD in Electrical Engineering and Computer Science at MIT under John N. Tsitsiklis.

2. 1997

   Temporal difference learning analysis

   With John Tsitsiklis, he coauthors “An Analysis of Temporal-Difference Learning with Function Approximation” in 1997. The paper appears in IEEE Transactions on Automatic Control.

3. 1998

   Joining the Stanford faculty

   In 1998, he joins Stanford University's faculty.

4. 2003

   Linear programming for decision processes

   In 2003, he and Daniela de Farias publish “The Linear Programming Approach to Approximate Dynamic Programming” in Operations Research. They also coauthor “Approximate Linear Programming for Average-Cost Dynamic Programming” at NeurIPS in 2003.

5. 2016

   An analysis of Thompson sampling

   In 2016, he and Daniel Russo publish “An Information-Theoretic Analysis of Thompson Sampling” in the Journal of Machine Learning Research. It analyzes learning from partial feedback in online optimization and relates performance bounds to uncertainty about the best action.

6. 2016

   Deep exploration with Bootstrapped DQN

   He coauthors “Deep Exploration via Bootstrapped DQN” with Ian Osband, Charles Blundell and Alexander Pritzel in 2016. The algorithm uses randomized value functions to explore and improves learning times and performance across most Atari games.

7. 2021–2023

   Epistemic neural networks

   In 2021, he coauthors the preprint “Epistemic Neural Networks,” which is revised in 2023. The work introduces the epinet, an addition to a conventional neural network that estimates uncertainty with modest extra computation.

8. 2022

   The Lanchester Prize

   In 2022, he and Daniel Russo win the INFORMS Frederick W. Lanchester Prize for their works compiled in “Learning to Optimize Via Posterior Sampling.” The work contributes to online statistical learning and optimization.

9. 2025

   The Morse Lectureship Award

   He receives the INFORMS Philip McCord Morse Lectureship Award in 2025 for his work connecting operations research and machine learning, including sequential decision making, reinforcement learning and alignment.

10. As of October 2026

    AI alignment at Stanford

    As of October 2026, he is a professor of Electrical Engineering and Management Science and Engineering at Stanford, with a courtesy appointment in Computer Science and membership in the Institute for Computational and Mathematical Engineering. His research focuses on reinforcement learning and alignment. His industry work includes co-founding Enuvis, working on Morgan Stanley's Process Driven Trading Team, joining Unica's founding team, and founding DeepMind's Efficient Agent Team, whose technology Google products use to adapt large language models to human feedback.