Sample-efficient Deep Reinforcement Learning Via Episodic Backward Update

Abstract

We propose Episodic Backward Update (EBU) - a novel deep reinforcement learning algorithm with a direct value propagation. In contrast to the conventional use of the experience replay with uniform random sampling, our agent samples a whole episode and successively propagates the value of a state to its previous states. Our computationally efficient recursive algorithm allows sparse and delayed rewards to propagate directly through all transitions of the sampled episode. We theoretically prove the convergence of the EBU method and experimentally demonstrate its performance in both deterministic and stochastic environments. Especially in 49 games of Atari 2600 domain, EBU achieves the same mean and median human normalized performance of DQN by using only 5% and 10% of samples, respectively.

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keylee2018sample

Related papers

• Episodic Reinforcement Learning With Expanded State-reward Space (2024)0.00
• Faster Deep Q-learning Using Neural Episodic Control (2018)0.00
• Sequential Memory Improves Sample And Memory Efficiency In Episodic Control (2021)0.00
• Sample-efficient Reinforcement Learning With Maximum Entropy Mellowmax Episodic Control (2019)0.00
• Sample Efficient Actor-critic With Experience Replay (2016)0.00
• Upside-down Reinforcement Learning Can Diverge In Stochastic Environments With Episodic Resets (2022)0.00
• Exploration Via Elliptical Episodic Bonuses (2022)3.58
• Rewarding Episodic Visitation Discrepancy For Exploration In Reinforcement Learning (2022)0.00