Sample-Efficient Learning of POMDPs with Multiple Observations In Hindsight

TL;DR

This paper introduces a new feedback model for POMDPs called 'multiple observations in hindsight' and demonstrates that it enables sample-efficient learning for two new subclasses, relaxing previous assumptions and broadening applicability.

Contribution

The paper proposes a novel feedback model and establishes sample-efficient learning algorithms for multi-observation revealing and distinguishable POMDPs, expanding the classes of POMDPs that can be learned efficiently.

Findings

Sample-efficient learning is achievable under the new feedback model.

Two new subclasses of POMDPs are identified for efficient learning.

Relaxed conditions for distinguishable POMDPs compared to revealing POMDPs.

Abstract

This paper studies the sample-efficiency of learning in Partially Observable Markov Decision Processes (POMDPs), a challenging problem in reinforcement learning that is known to be exponentially hard in the worst-case. Motivated by real-world settings such as loading in game playing, we propose an enhanced feedback model called ``multiple observations in hindsight'', where after each episode of interaction with the POMDP, the learner may collect multiple additional observations emitted from the encountered latent states, but may not observe the latent states themselves. We show that sample-efficient learning under this feedback model is possible for two new subclasses of POMDPs: \emph{multi-observation revealing POMDPs} and \emph{distinguishable POMDPs}. Both subclasses generalize and substantially relax \emph{revealing POMDPs} -- a widely studied subclass for which sample-efficient learning is possible under standard trajectory feedback. Notably, distinguishable POMDPs only require the emission distributions from different latent states to be \emph{different} instead of \emph{linearly independent} as required in revealing POMDPs.