一种面向种群训练的分布式深度强化学习训练模型
By designing a distributed deep reinforcement learning training model oriented towards population training and optimizing data exchange using a cache pool, the problems of uneven network bandwidth usage and insufficient agent interaction are solved, thus achieving efficient agent network parameter training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF AUTOMATION CHINESE ACAD OF SCI
- Filing Date
- 2022-10-25
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies suffer from uneven network bandwidth usage in distributed deep reinforcement learning training, resulting in low training efficiency. Furthermore, agents cannot interact during parallel training, failing to meet the needs of population training.
Design a distributed deep reinforcement learning training model for population training, which includes a training module, a data cache pool, a policy update module, and a weight cache pool. The cache pool alleviates network transmission pressure, optimizes the data exchange process, and supports training of a variable number of agent network parameters.
It significantly accelerates the network parameter training process of agents, solves the network congestion problem, realizes effective interaction and independent training between agents, and improves training efficiency.
Smart Images
Figure CN115496206B_ABST