ConfigX: Modular Configuration for Evolutionary Algorithms via Multitask Reinforcement Learning

TL;DR

ConfigX introduces a universal MetaBBO agent capable of configuring a broad space of evolutionary algorithms through Modular-BBO modularization and a Transformer-based multitask RL policy. By defining a joint optimization task space and leveraging module-aware attention, it learns a policy that generalizes to unseen EAs and problem instances, achieving strong zero-shot performance and efficient lifelong adaptation via fine-tuning. The approach demonstrates competitive to superior performance against SMAC3, with ablations highlighting the importance of architectural components such as MSA and positional encoding. This work advances automatic, all-purpose configuration for EAs, enabling scalable, data-driven EA design across diverse BBO tasks.

Abstract

Recent advances in Meta-learning for Black-Box Optimization (MetaBBO) have shown the potential of using neural networks to dynamically configure evolutionary algorithms (EAs), enhancing their performance and adaptability across various BBO instances. However, they are often tailored to a specific EA, which limits their generalizability and necessitates retraining or redesigns for different EAs and optimization problems. To address this limitation, we introduce ConfigX, a new paradigm of the MetaBBO framework that is capable of learning a universal configuration agent (model) for boosting diverse EAs. To achieve so, our ConfigX first leverages a novel modularization system that enables the flexible combination of various optimization sub-modules to generate diverse EAs during training. Additionally, we propose a Transformer-based neural network to meta-learn a universal configuration policy through multitask reinforcement learning across a designed joint optimization task space. Extensive experiments verify that, our ConfigX, after large-scale pre-training, achieves robust zero-shot generalization to unseen tasks and outperforms state-of-the-art baselines. Moreover, ConfigX exhibits strong lifelong learning capabilities, allowing efficient adaptation to new tasks through fine-tuning. Our proposed ConfigX represents a significant step toward an automatic, all-purpose configuration agent for EAs.

Figures (6)

• Figure 1: Conceptual overview of different AC paradigms.
• Figure 2: Left: The hierarchical polymorphism in Modular-BBO. Right: Legal/Illegal algorithm examples in Modular-BBO.
• Figure 3: The workflow of the Transformer based configuration policy in ConfigX.
• Figure 4: Optimization curves of the pre-trained ConfigX model and the baselines, over three different zero-shot scenarios.
• Figure 5: The learning curves of fine-tuning and re-training ConfigX on novel optimization problems or algorithm structures. The fine-tuning saves $3$x and $2$x learning steps than the re-training on $T_\text{test,out}^{(1)}$ and $T_\text{test,out}^{(2)}$ respectively.