RL via regressing relative values

TLDR: A tutorial on DPO/cDPO/IPO/RSO.

TLDR: How to use diversity search methods to obtain adversarial jailbreaks to policy?

TLDR: Bridging RL Theory and Practice.

TLDR: A novel framework for Q-learning that models the maximal soft-values without needing to sample from a policy.

rank-game: A unifying theory for learning from preferences and demonstrations.

TLDR: A explanation for the mutual information representation learning in RL.

TLDR: How can parallel and differentiable simulators speed up reinforcement learning?