UoE RL Reading Group | 13 January 2022

Speaker: Rishabh Agarwal (Google DeepMind)

Title: Deep reinforcement learning at the edge of the statistical precipice

Authors: Agarwal, Rishabh, Max Schwarzer, Pablo Samuel Castro, Aaron Courville, and Marc G. Bellemare

Abstract: I’ll talk about recent work, accepted at NeurIPS 2021 as an outstanding paper, where we find that statistical issues have a large influence on reported results on widely-used RL benchmarks. To help researchers do good and reliable science, I’ll discuss how to reliably evaluate and report performance on reinforcement learning (and ML?) benchmarks, especially when using only a handful of runs. See agarwl.github.io/rliable for details.

In: International Conference on Machine Learning, pp. 2388-2397. PMLR, 2021.

Link: https://arxiv.org/abs/2108.13264

Bio: Rishabh is a research scientist in the Google Brain team in Montréal. Previously, he was an AI Resident in Geoffrey Hinton’s team at Google Toronto. His research interests mainly revolve around deep reinforcement learning (RL), often with the goal of making RL methods suitable for real-world problems. https://agarwl.github.io