连接大脑与机器:神经科学、人工智能与神经形态系统的统一前沿
神经元与认知
2026-04-13 v2 神经与进化计算
摘要
这篇立场与综述论文识别了神经科学、通用人工智能(AGI)与神经形态计算向统一研究范式汇聚的新兴趋势。我们以基于大脑生理学的框架,重点阐述突触可塑性、稀疏脉冲通信和多模态关联如何为下一代AGI系统提供设计原则,这些系统有望融合人类智能与机器智能。本综述追溯了这一演化历程,从早期的连接主义模型到当前最先进的大语言模型,展示了诸如Transformer注意力、基础模型预训练和多智能体架构等关键创新如何镜像皮层机制、工作记忆和情景巩固等神经生物学过程。随后我们讨论了能够突破冯·诺依曼瓶颈以在硅基上实现大脑规模效率的新兴物理基底:忆阻交叉阵列、存内计算阵列以及新兴的量子与光子器件。在这一交叉领域存在四项关键挑战:1)将脉冲动力学与基础模型相融合;2)在不发生灾难性遗忘的前提下维持终身可塑性;3)在具身智能体中统一语言学习与感觉运动学习;4)在先进的神经形态自主系统中实施伦理保障。这一横跨神经科学、计算与硬件的综合视角,为上述各领域提供了一份整合性的研究议程。
引用
@article{arxiv.2507.10722,
title = {Bridging Brains and Machines: A Unified Frontier in Neuroscience, Artificial Intelligence, and Neuromorphic Systems},
author = {Sohan Shankar and Yi Pan and Hanqi Jiang and Zhengliang Liu and Mohammad R. Darbandi and Agustin Lorenzo and Junhao Chen and Weihang You and Md Mehedi Hasan and Arif Hassan Zidan and Eliana Gelman and Joshua A. Konfrst and Jillian Y. Russell and Katelyn Fernandes and Tianze Yang and Yiwei Li and Huaqin Zhao and Afrar Jahin and Triparna Ganguly and Shair Dinesha and Yifan Zhou and Zihao Wu and Xinliang Li and Lokesh Adusumilli and Aziza Hussein and Sagar Nookarapu and Jixin Hou and Kun Jiang and Jiaxi Li and Brenden Heinel and XianShen Xi and Hailey Hubbard and Zayna Khan and Levi Whitaker and Ivan Cao and Max Allgaier and Andrew Darby and Lin Zhao and Lu Zhang and Xiaoqiao Wang and Xiang Li and Wei Zhang and Xiaowei Yu and Dajiang Zhu and Yohannes Abate and Tianming Liu},
journal= {arXiv preprint arXiv:2507.10722},
year = {2026}
}