一种捕捉环境中可控因素的表示学习方法及系统
By introducing the concept of controllable factors and mutual information measurement, a representation learning method was designed to solve the problem of noise interference in reinforcement learning and improve the robustness and accuracy of policy training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF COMPUTING TECH CHINESE ACAD OF SCI
- Filing Date
- 2022-08-23
- Publication Date
- 2026-07-17
AI Technical Summary
Existing reinforcement learning algorithms struggle to effectively distinguish predictable noise in the environment when processing high-dimensional observation data, causing task-related information to be squeezed out in low-dimensional representations and affecting policy training performance.
By introducing the concept of controllable factors, a representation learning method is designed. The method maximizes the content of controllable factors using mutual information metric and loss function. Convolutional neural networks and multi-layer fully connected networks are used to construct the encoder to filter noise and improve the robustness of representation learning.
Effective filtering of predictable noise in the environment improves the robustness of representation learning and enhances the accuracy and efficiency of policy training.
Smart Images
Figure CN117688983B_ABST