DECN: Evolution Inspired Deep Convolution Network for Black-box Optimization

TL;DR

DECN introduces AutomatedEA by learning update rules for generation and selection within evolutionary search via a convolutional reasoning module and a selection module, forming end-to-end Evolution Modules. Trained with differentiable surrogate losses and surrogate function datasets, DECN achieves fast, task-tailored optimization that transfers to unseen objectives and scales with GPUs, outperforming state-of-the-art hand-designed and meta-learning EAs. The framework demonstrates strong generalization across synthetic, robotic, and deep-learning tasks, and analysis shows a learned exploration-to-exploitation progression. This approach offers a practical pathway to efficient, automated black-box optimization with broad applicability in architecture search, hyperparameter tuning, and control problems.

Abstract

Evolutionary algorithms (EAs) have emerged as a powerful framework for optimization, especially for black-box optimization. Existing evolutionary algorithms struggle to comprehend and effectively utilize task-specific information for adjusting their optimization strategies, leading to subpar performance on target tasks. Moreover, optimization strategies devised by experts tend to be highly biased. These challenges significantly impede the progress of the field of evolutionary computation. Therefore, this paper first introduces the concept of Automated EA: Automated EA exploits structure in the problem of interest to automatically generate update rules (optimization strategies) for generating and selecting potential solutions so that it can move a random population near the optimal solution. However, current EAs cannot achieve this goal due to the poor representation of the optimization strategy and the weak interaction between the optimization strategy and the target task. We design a deep evolutionary convolution network (DECN) to realize the move from hand-designed EAs to automated EAs without manual interventions. DECN has high adaptability to the target task and can obtain better solutions with less computational cost. DECN is also able to effectively utilize the low-fidelity information of the target task to form an efficient optimization strategy. The experiments on nine synthetics and two real-world cases show the advantages of learned optimization strategies over the state-of-the-art human-designed and meta-learning EA baselines. In addition, due to the tensorization of the operations, DECN is friendly to the acceleration provided by GPUs and runs 102 times faster than EA.

Figures (9)

• Figure 1: A general view of DECN and EM.
• Figure 2: The compared results on high-fidelity surrogate functions. (a)-(f) represent the convergence curves of DECN and the comparison algorithms on F4-F9 when $D$=10.
• Figure 3: The compared results on high-fidelity surrogate functions. (a)-(f) represent the convergence curves of DECN and the comparison algorithms on F4-F9 when $D$=100.
• Figure 4: The compared results of different DECN structures trained based on low-fidelity surrogate functions. (a)-(f) represent the convergence curves of DECN and the comparison algorithms on F4-F9 when $D$=10.
• Figure 5: The compared results of different DECN structures trained based on low-fidelity surrogate functions. (a)-(f) represent the convergence curves of DECN and the comparison algorithms when $D$=100.