A Theoretical Understanding of Gradient Bias in Meta-Reinforcement Learning
Nov 28, 2022
Speakers
About
Gradient-based Meta-RL (GMRL) refers to methods that maintain two-level optimisation procedures wherein the outer-loop meta-learner guides the inner-loop gradient-based reinforcement learner to achieve fast adaptations. In this paper, we develop a unified framework that describes variations of GMRL algorithms and points out that existing stochastic meta-gradient estimators adopted by GMRL are actually biased. Such meta-gradient bias comes from two sources: 1) the compositional bias incurred by the two-level problem structure, which has an upper bound of 𝒪(Kα^Kσ̂_In|τ|^-0.5) w.r.t. inner-loop update step K, learning rate α, estimate variance σ̂^2_In and sample size |τ|, and 2) the multi-step Hessian estimation bias Δ̂_H due to the use of autodiff, which has a polynomial impact 𝒪((K-1)(Δ̂_H)^K-1) on the meta-gradient bias. We study tabular MDPs empirically and offer quantitative evidence that testifies our theoretical findings on existing stochastic meta-gradient estimators. Furthermore, we conduct experiments on Iterated Prisoner's Dilemma and Atari games to show how other methods such as off-policy learning and low-bias estimator can help fix the gradient bias for GMRL algorithms in general.Gradient-based Meta-RL (GMRL) refers to methods that maintain two-level optimisation procedures wherein the outer-loop meta-learner guides the inner-loop gradient-based reinforcement learner to achieve fast adaptations. In this paper, we develop a unified framework that describes variations of GMRL algorithms and points out that existing stochastic meta-gradient estimators adopted by GMRL are actually biased. Such meta-gradient bias comes from two sources: 1) the compositional bias incurred by t…
Organizer
NeurIPS 2022
Account · 960 followers
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
MinVIS: A Minimal Video Instance Segmentation Framework without Video-based Training
Watch later
De-An Huang, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Vision-centric Autonomous Driving: from Perception to Prediction
Watch later
Total of 1 viewers voted for saving the presentation to eternal vault which is 0.1%
First Steps Toward Understanding the Extrapolation of Nonlinear Models to Unseen Domains
Watch later
Kefan Dong, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Interactive Imitation Learning in Robotics
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Language Models Can Teach Themselves to Program Better
Watch later
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Cache-Augmented Inbatch Importance Resampling for Training Recommender Retriever
Watch later
Jin Chen, …
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%