Deep Double Q-learning Reduces Overestimation Bias in Reinforcement Learning

Researchers have introduced Deep Double Q-learning (DDQL), an advancement over Double DQN designed to mitigate overestimation bias in deep reinforcement learning. DDQL explicitly trains two Q-functions, unlike Double DQN which maintains correlated estimators. The new algorithm employs techniques such as lower replay ratios and longer target network update intervals to stabilize training. In evaluations across 57 Atari 2600 games, DDQL demonstrated improved aggregate performance, outperforming Double DQN on 47 games and further reducing overestimation. AI

IMPACT This research could lead to more stable and accurate reinforcement learning agents, improving performance in complex control tasks.

RANK_REASON The cluster describes a new algorithm presented in a research paper on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Atari 2600
• Deep Double Q-learning
• Double Q-learning
• Prabhat Nagarajan
• Q-learning

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →