Memory Merge DQN enhances stable value learning in Atari games

Researchers have introduced Memory Merge DQN, a novel target network update mechanism for deep Q-networks designed to improve the stability and final performance of agents in reinforcement learning. This method constructs the target network by merging recent historical copies of the online network, weighting parameters based on their influence on Q-value sensitivity rather than simply copying the newest parameters. This approach aims to preserve useful value function structure during long-horizon learning. Evaluations on Atari environments demonstrated that Memory Merge DQN is competitive, achieving superior performance compared to standard DQN, Averaged DQN, and PQN in several games. AI

IMPACT This novel target network update mechanism could lead to more stable and higher-performing reinforcement learning agents in complex environments.

RANK_REASON The cluster contains a research paper detailing a new method for deep Q-networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Atari
• Averaged DQN
• Fisher Weight Model Merging
• Memory Merge DQN

• paper
• other

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