A Multiobjective Evolutionary Algorithm Based on Bipopulation With Uniform Sampling for Neural Architecture Search.

2026-02-09

IEEE Transactions on Neural Networks and Learning Systems

AI Summary

This work addresses the challenges in multi-objective neural architecture search, where insufficient population diversity and inadequate exploration of the search space often hinder the simultaneous optimization of accuracy and network complexity. To overcome these limitations, the authors propose MOEA-BUS, a novel algorithm that integrates uniform-sampling initialization, a dual-population co-evolutionary mechanism, and a support vector machine surrogate model. This approach significantly enhances population diversity and improves exploration in regions of extreme computational complexity. Evaluated on CIFAR-10 and ImageNet, the method achieves top-1 accuracies of 98.39% and 80.03%, respectively. Notably, it attains 78.28% accuracy on ImageNet with only 446M MAdds, while improving the Kendall’s tau correlation coefficient by approximately 0.07, substantially outperforming existing state-of-the-art methods.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Neural architecture search (NAS) automates neural network design, improving efficiency over manual approaches. However, efficiently discovering high-performance neural network architectures that simultaneously optimize multiple objectives remains a significant challenge in NAS. Existing methods often suffer from limited population diversity and inadequate exploration of the search space, particularly in regions with extreme complexity values. To address these challenges, we propose a multiobjective evolutionary algorithm (MOEA)-BUS, an innovative MOEA based on bipopulation with uniform sampling for NAS, aimed at simultaneously optimizing both accuracy and network complexity. In MOEA-BUS, a novel uniform sampling method is proposed to initialize the population, ensuring that architectures are distributed uniformly across the objective space. Furthermore, to enhance exploration, we deploy a bipopulation framework where two populations evolve synergistically, facilitating comprehensive search space coverage. Experiments on CIFAR-10 and ImageNet demonstrate MOEA-BUS's superiority, achieving top-one accuracies of 98.39% on CIFAR-10, and 80.03% on ImageNet. Notably, it achieves 78.28% accuracy on ImageNet with only 446 M MAdds. Ablation studies confirm that both uniform sampling and bipopulation mechanisms enhance population diversity and performance. In addition, in terms of Kendall's tau coefficient, the SVM achieves an improvement of at least 0.035 compared to the other three commonly used machine learning models, and uniform sampling provided an enhancement of approximately 0.07.

Problem

• Neural Architecture Search
• Multi-objective Optimization
• Population Diversity
• Search Space Exploration
• Network Complexity

Innovation

• Neural Architecture Search
• Multi-objective Evolutionary Algorithm
• Bi-population
• Uniform Sampling
• Population Diversity