基于离体神经网络的生物智能计算:关键技术与研究现状(特邀)  

Biological intelligent computing based on in vitro neural networks:Key technologies and research status(invited)

作　　者：杜金帅 邓寅 曹世阳 陆泽营 李杰[1] 韩卓 周晋民 王棵 桂丽丽 徐坤[1] DU Jinshuai;DENG Yin;CAO Shiyang;LU Zeying;LI Jie;HAN Zhuo;ZHOU Jinmin;WANG Ke;GUI Lili;XU Kun(National Key Laboratory of Information Photonics and Optical Communications,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学信息光子学与光通信全国重点实验室,北京100876

出　　处：《红外与激光工程》2025年第3期93-113,I0001,共22页Infrared and Laser Engineering

基　　金：国家自然科学基金项目(62401083);中央高校基本科研业务费专项基金项目(ZDYY202102);北京市自然科学基金项目(4232010);信息光子学与光通信全国重点实验室(北京邮电大学)基金项目(IPOC2021ZR02)。

摘　　要：随着人工智能(AI)的快速发展,其高能耗、低解释性和灵活性不足等问题日益凸显。生物神经网络(Biological Neural Networks,BNN)因其基于真实生物神经元的动态可塑性和自适应性,成为突破AI瓶颈的重要研究方向。离体培养的BNN通过精确的培养技术和先进的调控手段,为理解生物大脑的信息处理机制以及开发新型智能系统提供了实验平台。文中系统地综述了离体BNN的培养方法及功能优化技术,包括二维和三维培养方式,以及模块化网络构建与连接调控方法。详细探讨了电学刺激、光学刺激和化学刺激等信号输入技术,以及细胞内外动作电位记录、钙荧光成像等信号输出技术。此外,还分析了BNN在静态任务(如语音识别)和实时交互任务(如神经机器人控制)中的应用实例,展示了其在动态学习、复杂模式识别和实时任务适应中的潜力。最后,总结了当前基于BNN的生物智能计算研究的主要挑战,提出了未来研究的方向,希望为未来生物智能计算的研究与发展提供一定启示。Significance Although current artificial intelligence(AI)systems,especially deep learning models,have made significant progress in multiple fields,they still face problems such as high energy consumption,low interpretability,and insufficient flexibility.In contrast,in vitro Biological Neural Networks(BNN)are based on real biological neurons.Through their unique dynamic plasticity and adaptive learning capabilities,they can complete complex information processing tasks with low power consumption,laying the foundation for AI.Energy efficiency,interpretability and flexibility lead to completely new solutions.In addition,in vitro BNN uses precise culture technology and advanced control methods to enable the function and structure of neural networks to be precisely controlled,thereby providing an experimental platform for understanding the working mechanism of the brain,optimizing intelligent computing systems,and advancing brain-computer interface technology.Although the current computing performance of BNN is still in its infancy,its characteristics provide important directions and ideas for the development of efficient intelligent computing systems in the future.Progress In recent years,bio-intelligent computing based on in vitro BNN has made significant progress in culture technology,signal monitoring and application scenarios.Through breakthroughs in two-dimensional and three-dimensional culture technology,BNN can more realistically simulate the biological brain neural network structure and promote the optimization of neural network functions.At the same time,combining multiple input methods such as electrical,optical and chemical stimulation allows neuronal activity to be precisely controlled,providing a new experimental platform for task adaptation and complex information processing.In terms of applications,BNN has shown great potential in static tasks(such as speech recognition)and real-time interactive tasks(such as neurorobot control),especially in dynamic learning,complex pattern recognition,and real-time t

关 键 词：生物神经网络 离体培养 双向通信技术 生物智能计算 

分 类 号：R318.01[医药卫生—生物医学工程]