Human brain is the product of evolution during hundreds over millions of years and can engage in multiple advanced cognitive functions with low energy consumption. Brain-inspired artificial intelligence serves as a computational continuation of this natural evolutionary process, is imperative to take inspiration from the evolutionary mechanisms of brain structure and function. Studies suggest that the human brain's high efficiency and low energy consumption may be closely related to its small-world topology and critical dynamics. However, existing efforts on the performance-oriented structural evolution of spiking neural networks (SNNs) are time-consuming and ignore the core structural properties of the brain. In this paper, we propose a multi-objective Evolutionary Liquid State Machine (ELSM) with the combination of small-world coefficient and criticality as evolution goals and simultaneously integrate the topological properties of spiking neural networks from static and dynamic perspectives to guide the emergence of brain-inspired efficient structures. Extensive experiments show a consistent and comparable performance of the proposed model compared to LSM-based and hierarchical SNNs algorithms: it achieves 97.23\% on NMNIST, and reaches the state-of-art performance compared to all LSM models on MNIST and Fashion-MNIST (98.05\% and 88.81\%, respectively). A thorough analysis reveals the spontaneous emergence of hub nodes, short paths, long-tailed degree distributions, and numerous community structures in evolutionary models. This work evolves recurrent spiking neural networks into brain-inspired efficient structures and dynamics, providing the potential to achieve adaptive general aritficial intelligence.

翻译：人脑是数亿年进化的产物，能够以极低能耗执行多种高级认知功能。脑启发的人工智能作为这一自然进化过程的计算延续，必须从脑结构与功能的进化机制中汲取灵感。研究表明，人脑的高效性与低能耗可能与其小世界拓扑结构及临界动力学特性密切相关。然而，现有面向性能的脉冲神经网络（SNN）结构进化方法不仅耗时，而且忽视了脑的核心结构特性。本文提出一种多目标进化液体状态机（ELSM），将小世界系数与临界性作为进化目标，并同时从静态与动态视角整合脉冲神经网络的拓扑特性，以引导脑启发高效结构的涌现。大量实验表明，与基于LSM及层次化SNN算法相比，所提模型性能一致且可比：在NMNIST数据集上达到97.23%，在MNIST和Fashion-MNIST数据集上分别达到98.05%和88.81%，超越所有LSM模型取得最优结果。深入分析揭示了进化模型中枢纽节点、短路径、长尾度分布及丰富社区结构的自发涌现。本研究将循环脉冲神经网络进化为脑启发的高效结构与动力学，为实现自适应通用人工智能提供了潜力。