Reinforcement Learning: Theory and Algorithms

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Is Reinforcement Learning (Not) for Natural Language Processing?: Benchmarks, Baselines, and Building Blocks for Natural Language Policy Optimization (2022, background): “, 2022) and the cliff MDP problem (Huszár, 2015; Agarwal et al., 2019; Swamy et al., 2021).”
Offline RL Without Off-Policy Evaluation (2021, background): “We don’t expect this to be easy as it requires understanding policy iteration which has been notoriously difﬁcult to analyze, often converging much faster than the theory would suggest [Sutton and Barto, 2018, Agarwal et al., 2019].”
COMBO: Conservative Offline Model-Based Policy Optimization (2021, methodology): “…third equality, we utilize the result on the difference between returns of a policy π on two different MDPs, P M 1 and P M f from Agarwal et al. [1] (Chapter 2, Lemma 2.2, Simulation Lemma), and additionally incorporate the auxiliary MDP M in the expression via addition and subtraction in the…”
VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training (2022, background): “First, we note that A proof can be found in Section 1.1.3 of Agarwal et al. (2019).”
Is Pessimism Provably Efficient for Offline RL? (2020, background): “In practice, such assumptions on the sufﬁcient coverage of the dataset often fail to hold (Fujimoto et al., 2019; Agarwal et al., 2020b; Fu et al., 2020a; Gulcehre et al., 2020), which possibly invalidates existing theory.”