Advancing CMA-ES with Learning-Based Cooperative Coevolution for Scalable Optimization

2025-04-24

AI Summary

Cooperative Coevolution (CC) for large-scale global optimization suffers from poor generalizability due to its reliance on manually designed variable decomposition strategies. Method: This paper proposes Learning-driven Cooperative Coevolution (LCC), a dynamic decomposition scheduling framework that introduces the first neural-network-based decomposition strategy selector, trained via Proximal Policy Optimization (PPO) reinforcement learning to enable cross-problem adaptive scheduling without prior knowledge. LCC integrates CMA-ES, meta black-box optimization, graph- and state-based feature engineering, and CC mechanisms. Contribution/Results: On diverse large-scale benchmark suites, LCC significantly outperforms state-of-the-art methods in convergence accuracy and computational efficiency, while demonstrating strong transferability across problems. These results validate the effectiveness and universality of the data-driven decomposition scheduling paradigm.

Technology Category

• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Recent research in Cooperative Coevolution~(CC) have achieved promising progress in solving large-scale global optimization problems. However, existing CC paradigms have a primary limitation in that they require deep expertise for selecting or designing effective variable decomposition strategies. Inspired by advancements in Meta-Black-Box Optimization, this paper introduces LCC, a pioneering learning-based cooperative coevolution framework that dynamically schedules decomposition strategies during optimization processes. The decomposition strategy selector is parameterized through a neural network, which processes a meticulously crafted set of optimization status features to determine the optimal strategy for each optimization step. The network is trained via the Proximal Policy Optimization method in a reinforcement learning manner across a collection of representative problems, aiming to maximize the expected optimization performance. Extensive experimental results demonstrate that LCC not only offers certain advantages over state-of-the-art baselines in terms of optimization effectiveness and resource consumption, but it also exhibits promising transferability towards unseen problems.

Problem

• Dynamic selection of decomposition strategies for large-scale optimization
• Reducing expertise dependency in cooperative coevolution frameworks
• Enhancing optimization performance and transferability via learning-based methods

Innovation

• Learning-based cooperative coevolution framework
• Neural network for dynamic decomposition strategy
• Reinforcement learning for optimization performance