This work proposes a middle ground in the form of a scalable information-theoretic generalization of CCA, termed max-sliced mutual information (mSMI).

TRPO, PPO), and a critic that involves minimizing a closely connected objective.

This phenomenon cannot be easily explained by offline RL's return maximization objective.

Both challenges involve modeling long-term dependencies.

In such situations, the successor representation (SR) is a popular framework which supports rapid policy evaluation by decoupling a policy's

Finding tight linear bounds for activation functions in neural networks is an essential part of several state of the art neural network robustness certification tools.