A priority experience replay actor-critic algorithm using self-attention mechanism for strategy optimization of...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Reinforcement Learning

Background:

• Self-attention mechanisms are increasingly recognized in deep reinforcement learning.
• Application of self-attention in discrete problem domains is limited due to optimization challenges.

Purpose of the Study:

• Introduce a novel deep reinforcement learning algorithm, Attention-based Actor-Critic with Priority Experience Replay (A2CPER).
• Enhance policy formulation for discrete problems by combining self-attention, Actor-Critic, and prioritized experience replay.

Main Methods:

• A2CPER utilizes dual networks (Actor and Critic) within the Actor-Critic framework.
• Incorporates target networks for stable optimization and self-attention mechanisms to focus on critical information.
• Employs priority experience replay to improve training stability and reduce sample correlation.

Main Results:

• Empirical experiments on discrete action problems demonstrate A2CPER's effectiveness in policy optimization.
• The algorithm achieved significant performance improvements across various tasks.
• A2CPER validates the viability of self-attention mechanisms in reinforcement learning for discrete problems.

Conclusions:

• A2CPER presents a robust framework for discrete problem-solving in deep reinforcement learning.
• The integration of self-attention mechanisms shows promise for complex decision-making scenarios.
• This approach offers potential applicability in advanced AI applications requiring efficient policy formulation.