Adam White

University of Alberta, Google DeepMind (United Kingdom)

Papers

Total Citations

H-Index

About

Adam White is a researcher specializing in reinforcement learning, continual prediction, and robot learning, with a particular focus on how intelligent systems can build rich, adaptive knowledge of their environments through ongoing sensorimotor interaction. His most celebrated work, "Multi-timescale Nexting in a Reinforcement Learning Robot" (2014, 68 citations), draws on psychological concepts to explore how robots can continuously predict near-future events across multiple timescales — a capability White frames as foundational to genuine environmental awareness. This contribution helped bridge cognitive science and machine learning in a meaningful way. White has also made important strides in scalable AI architectures, contributing to the development of Horde, a system enabling robots to learn diverse knowledge representations through unsupervised interaction. His more recent investigations into step-size adaptation methods for non-stationary prediction problems and rigorous empirical comparisons of off-policy learning algorithms reflect a commitment to making continual learning practical and robust. Collectively, his work advances the ambitious goal of building machines that learn and predict like living organisms — dynamically, persistently, and intelligently — positioning him as a thoughtful contributor to the frontier of real-world reinforcement learning research.