Tom Blau

Papers

Total Citations

H-Index

About

Tom Blau is a researcher advancing the frontier of reinforcement learning (RL), with a focus on making autonomous agents more sample-efficient and intelligent in their exploration strategies. His key research areas include Bayesian deep learning, curiosity-driven exploration, and pre-training techniques for RL. Blau’s major contributions address two critical bottlenecks in modern RL: the high sample complexity of learning control policies and the inefficiency of naive exploration methods like ε-greedy. In his 2018 work, "Improving Reinforcement Learning Pre-Training with Variational Dropout," he introduced a method to enhance pre-training for robotic control tasks—such as maze navigation and bipedal locomotion—by leveraging variational dropout to improve generalization and reduce wasted training steps. His 2019 paper, "Bayesian Curiosity for Efficient Exploration in Reinforcement Learning," proposed a Bayesian approach to curiosity, enabling agents to systematically prioritize unexplored or uncertain states, thereby accelerating learning in complex environments. Both papers have garnered 7 citations each, reflecting early but meaningful impact in the RL community. Blau’s work is particularly notable for bridging Bayesian inference with practical RL algorithms, offering principled solutions to the exploration-exploitation dilemma. For students and researchers, his contributions provide a clear pathway toward more autonomous, efficient, and curious learning agents.