Transfer from Multiple Linear Predictive State Representations (PSR)

TL;DR

This paper introduces a novel approach for transferring policies between partially observable environments using multiple linear predictive state representations, addressing a gap in transfer learning for partial observability.

Contribution

It presents the first algorithms for policy transfer in partially observable settings using PSRs, with detailed analysis of their performance and limitations.

Findings

Algorithms successfully transfer policies with known models

Performance analysis highlights strengths and weaknesses

Lays groundwork for future transfer learning research in POMDPs

Abstract

In this paper, we tackle the problem of transferring policy from multiple partially observable source environments to a partially observable target environment modeled as predictive state representation. This is an entirely new approach with no previous work, other than the case of transfer in fully observable domains. We develop algorithms to successfully achieve policy transfer when we have the model of both the source and target tasks and discuss in detail their performance and shortcomings. These algorithms could be a starting point for the field of transfer learning in partial observability.