Nicolas Perrin-Gilbert - Off-policy RL and interactive omics data exploration

The podcast features Nicolas Perron Gilbert, a CNRS researcher specializing in learning algorithms for robotics, particularly reinforcement learning, who discusses two main topics: the AFU algorithm for off-policy reinforcement learning and the Yomics tool for interactive exploration and analysis of omics data. Gilbert explains the importance of off-policy reinforcement learning in robotics due to its sample efficiency, especially in continuous action spaces. He introduces AFU as an improvement over Implicit Q-Learning (IQL), making it efficient for both offline and online reinforcement learning by using adaptive regularization to balance speed and precision. Additionally, Gilbert presents YoMix, a tool developed with the Curie Institute, designed to assist bioinformaticians in analyzing large-scale RNA sequencing data through interactive data visualization and gene signature identification.