LGFeb 16, 2021

Model-based Meta Reinforcement Learning using Graph Structured Surrogate Models

arXiv:2102.08291v127 citations
Originality Highly original
AI Analysis

This work addresses bottlenecks in real-world reinforcement learning applications by improving data efficiency and generalization across tasks, representing a novel method for a known bottleneck.

The paper tackles low data efficiency and weak generalization in reinforcement learning by proposing a Thompson-sampling based approach with a graph structured surrogate model (GSSM) for predicting environment dynamics, which outperforms state-of-the-art methods and achieves high returns while enabling fast execution by avoiding test-time policy optimization.

Reinforcement learning is a promising paradigm for solving sequential decision-making problems, but low data efficiency and weak generalization across tasks are bottlenecks in real-world applications. Model-based meta reinforcement learning addresses these issues by learning dynamics and leveraging knowledge from prior experience. In this paper, we take a closer look at this framework, and propose a new Thompson-sampling based approach that consists of a new model to identify task dynamics together with an amortized policy optimization step. We show that our model, called a graph structured surrogate model (GSSM), outperforms state-of-the-art methods in predicting environment dynamics. Additionally, our approach is able to obtain high returns, while allowing fast execution during deployment by avoiding test time policy gradient optimization.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes