Ferenc Huszár • 11/12/2020

Notes on Causally Correct Partial Models

This article provides a detailed breakdown of the paper 'Causally Correct Partial Models for Reinforcement Learning' by Rezende et al. (2020). It explains the Partially Observed Markov Decision Process (POMDP) setup, the agent's goal of maximizing reward, and the core challenge of estimating the expected performance of a new policy through causal/counterfactual queries, contrasting model-based and model-free approaches like importance sampling.

0 comments

#Machine Learning #Reinforcement Learning #Causal Inference