Recently, researchers have found that representations learned by large-scale pre-trained language models are useful in various downstream tasks.

We introduce a principled method for performing zero-shot transfer in reinforcement learning (RL) by exploiting approximate models of the environment. Zero-shot transfer in RL has been investigated by leveraging methods rooted in generalized policy improvement (GPI) and successor features (SFs).

The conditions on the representation that imply global convergence are different between these two algorithms.

Adjusted Mutual Information, are type II biased.

Müller et al. [2021], Nguyen et al. [2022] builds a GP model for both the function value and the