Wang Xuebin, Liu Chunxiuzi, Zhao Meng, Zhang Ke, Di Zengru, Liu He
Department of Systems Science, Faculty of Arts and Sciences, Beijing Normal University, Zhuhai, 519087, China.
International Academic Center of Complex Systems, Beijing Normal University, Zhuhai, 519087, China.
Adv Sci (Weinh). 2025 Feb;12(7):e2410637. doi: 10.1002/advs.202410637. Epub 2024 Dec 16.
This study introduces an artificial neural network (ANN) for image classification task, inspired by the aversive olfactory learning neural circuit in Caenorhabditis elegans (C. elegans). Although artificial neural networks (ANNs) have demonstrated remarkable performance in various tasks, they still encounter challenges including excessive parameterization, high training costs and limited generalization capabilities, etc. C. elegans, boasting a simple nervous system consisting of merely 302 neurons, is capable of exhibiting complex behaviors such as aversive olfactory learning. This research pinpoints key neural circuit related to aversive olfactory learning in C. elegans by means of behavioral experiment and high-throughput RNA sequencing, and then translates it into an architecture of ANN for image classification. Furthermore, other ANNs for image classification with different architectures are constructed for comparative performance analysis to underscore the advantages of the bio-inspired designed architecture. The results show that the ANN inspired by the aversive olfactory learning neural circuit in C. elegans attains higher accuracy, greater consistency and faster convergence rate in the image classification task, particularly when dealing with more complex classification challenges. This study not only demonstrates the potential of bio-inspired design in improving the capabilities of ANNs but also offers a novel perspective and methodology for future ANNs design.
本研究受秀丽隐杆线虫（C. elegans）厌恶嗅觉学习神经回路的启发，引入了一种用于图像分类任务的人工神经网络（ANN）。尽管人工神经网络（ANNs）在各种任务中都表现出了卓越的性能，但它们仍然面临诸多挑战，包括参数化过多、训练成本高以及泛化能力有限等。秀丽隐杆线虫拥有仅由302个神经元组成的简单神经系统，却能够展现出诸如厌恶嗅觉学习等复杂行为。本研究通过行为实验和高通量RNA测序，确定了秀丽隐杆线虫中与厌恶嗅觉学习相关的关键神经回路，然后将其转化为用于图像分类的ANN架构。此外，还构建了具有不同架构的其他用于图像分类的ANNs进行性能对比分析，以突出受生物启发设计的架构的优势。结果表明，受秀丽隐杆线虫厌恶嗅觉学习神经回路启发的ANN在图像分类任务中达到了更高的准确率、更好的一致性和更快的收敛速度，尤其是在处理更复杂的分类挑战时。本研究不仅证明了生物启发设计在提高ANNs能力方面的潜力，还为未来ANNs的设计提供了新的视角和方法。
