Off-Policy Actor-Critic with Shared Experience Replay

By Simon Schmitt, Matteo Hessel, Karen Simonyan. 2019

71 citations recorded by the source corpus.

We investigate the combination of actor-critic reinforcement learning algorithms with uniform large-scale experience replay and propose solutions for two challenges: (a) efficient actor-critic learning with experience replay (b) stability of off-policy learning where agents learn from other agents behaviour. We employ those insights to accelerate hyper-parameter sweeps in which all participating agents run concurrently and share their experience via a common replay module. To this end we analyze the bias-variance tradeoffs in V-trace, a form of importance sampling for actor-critic methods. Based on our analysis, we then argue for mixing experience sampled from replay with on-policy experience, and propose a new trust region scheme that scales effectively to data distributions where V-trace becomes unstable. We provide extensive empirical validation of the proposed solution. We further show the benefits of this setup by demonstrating state-of-the-art data efficiency on Atari among agents trained up until 200M environment frames.

Subfields: Artificial Intelligence, Management Science and Operations Research, Reinforcement Learning

How researchers see this paper

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

In Artificial Intelligence

In Management Science and Operations Research

In Reinforcement Learning

Citations in the corpus

Mastering Atari, Go, chess and shogi by planning with a learned model (2019, methodology): “When applied to 57 Atari games, using 200 million frames of experience per game, MuZero Reanalyze achieved 731% median normalized score, compared to 192%, 231% and 431% for previous state-of-the-art model-free approaches IMPALA [9], Rainbow [17] and LASER [36] respectively.”
V-MPO: On-Policy Maximum a Posteriori Policy Optimization for Discrete and Continuous Control (2019, methodology): “The ResNet used to process visual observations is similar to the 3-section ResNet used in Hessel et al. (2018), except the number of channels was multiplied by 4 in each section, so that the number of channels were (64, 128, 128) (Schmitt et al., 2019).”
Self-Tuning Deep Reinforcement Learning (2020, background): “In fact, there are only two agents that reported better performance after 200M frames: LASER (Schmitt et al., 2019) achieved a normalized median score of 431% and MuZero (Schrittwieser et al., 2019) achieved 731%.”
Mastering Diverse Domains through World Models (2023, methodology): “Experience replay We implement Dreamer using a uniform replay buffer with an online queue 57 .”
Mastering diverse control tasks through world models (2025, unknown): “We implement Dreamer using a uniform replay buffer with an online queue 63 .”
Bootstrapped Meta-Learning (2021, methodology): “We add experience replay (as per (Schmitt et al., 2020)) to allow multiple steps on the target.”
Experience Replay with Likelihood-free Importance Weights (2020, background): “The ﬁrst source is the evaluation of the TD learning target for longer traces such as TD( λ ); importance weights can be used to debias targets computed from off-policy trajectories [30, 26, 11, 33], similar to its role in policy learning.”
Sample Factory: Egocentric 3D Control from Pixels at 100000 FPS with Asynchronous Reinforcement Learning (2020, methodology): “Hybrid methods such as LASER (Schmitt et al., 2019) promise to combine high scalability, flexibility, and sample efficiency.”