2024 CEC Special Session

Evolutionary Reinfocement Learning, a new way to rethink exploration vs exploitation.

Introduction

Deep reinforcement learning has achieved remarkable results in Go, robotics, large language models, games, and other fields. However, problems such as sensitivity to hyperparameters, credit assignment for sparse rewards and multi-agent rewards, diversity exploration for deceptive rewards or continuous state and action spaces, and training efficiency pose great challenges to deep reinforcement learning. In recent years, a lot of studies combining evolutionary computation and reinforcement learning have been proposed, namely evolutionary reinforcement learning, to improve the performance of reinforcement learning in terms of policy diversity, action novelty, and robustness to tasks with sparse or delayed rewards, etc. Therefore, evolutionary reinforcement learning is a promising research direction. This special session “Evolutionary Reinforcement Learning and Applications” is to encourage further research on combination mechanisms of evolutionary reinforcement learning, parallel and distributed computing methods to improve training efficiency, applications to real-world problems, etc. Evolutionary reinforcement learning is a cross-fertilization of two technique disciplines of WCCI, i.e., evolutionary computation and neural networks. It is beneficial for evolutionary computation to play a greater role in the new generation of artificial intelligence.

The purpose of this special session on “Evolutionary Reinforcement Learning and Applications” is to provide a forum for researchers in the field to develop advanced evolutionary reinforcement learning techniques and apply them to various real-world scenarios.

Scope

The topics of this special session include but are not limited to the following topics:

Theory study of evolutionary reinforcement learning
- Sample efficiency of evolutionary reinforcement learning
- Diversity encouragement of evolutionary reinforcement learning
- Novelty seeking of evolutionary reinforcement learning
- Reward shaping of evolutionary reinforcement learning
- Action selection of evolutionary reinforcement learning
- Policy search based on genetic algorithms
- Policy search based on evolution strategies
- Policy search based on genetic programming
- Parallel and distributed evolutionary reinforcement learning
Novel combination of evolutionary computation and reinforcement learning
- Evolutionary computation for meta-reinforcement learning
- Evolutionary computation for multi-objective reinforcement learning
- Evolutionary computation for multi-agent reinforcement learning
- Evolutionary computation for hyperparameter optimization
- Evolutionary computation for generalized reinforcement learning
Application study of evolutionary reinforcement learning
- Intelligent transportation systems
- Unmanned aerial vehicles
- Robot control and scheduling
- Natural Language Processing
- Supply Chain and Logistics
- Cloud computing and edge computing
- Realtime multi-agent Game
- Recommendation systems