正文：
近日，中国科学院自动化研究所的研究团队在人工智能与神经科学交叉领域取得重要进展。该团队提出的基于内生复杂性的类脑网络模型，被发表于《Nature Computational Science》期刊上。该研究成果为人工智能领域的发展提供了一种新的思路，旨在构建更加广泛和通用的认知能力。
该研究团队的共同通讯作者包括中国科学院自动化所的李国齐研究员和徐波研究员，以及北京大学田永鸿教授。共同第一作者为清华大学钱学森班的本科生何林轩、数理基科班的本科生徐蕴辉，以及清华大学精仪系的博士生何炜华和林逸晗。
当前人工智能领域的发展主要依赖于“基于外生复杂性”的方法，即通过构建更大、更深和更宽的神经网络来提升模型的表现。然而，这种方法面临着高昂的计算资源和能源消耗，以及可解释性不足等问题。为了克服这些困境，研究团队借鉴了大脑神经元的复杂动力学特性，提出了基于内生复杂性的类脑神经元模型构建方法。
研究团队提出的“具有内生复杂性的小网络模型”概念，通过模拟生物神经元的复杂动力学，将复杂的内部结构引入单个神经元，从而构建更高效的AI模型。研究者在脉冲神经网络中使用了HH模型来替代传统的LIF模型，这是一种描述神经元动作电位产生机制的数学模型，能够模拟神经元对各种刺激的响应。
通过理论证明和仿真研究，研究团队展示了HH模型与LIF模型在动作电位产生机制上的等效性，以及如何通过设计微结构提升计算单元的内生复杂性。这项研究为人工智能领域提供了一种新的模型构建方法，有望在未来的AI研究中发挥重要作用。
该研究成果不仅为人工智能领域的发展提供了新的理论基础和技术路径，也为神经科学的研究提供了新的视角和方法。随着人工智能与神经科学的深入合作，未来有望在认知科学、神经疾病治疗等领域取得更多突破。
英语如下：
News Title: “New Brain-like Network: A New Milestone in Cross-disciplinary Collaboration between Artificial Intelligence and Neuroscience”
Keywords: Brain Network, Artificial Intelligence, Cognitive Capabilities
News Content:
Recently, a research team from the Institute of Automation, Chinese Academy of Sciences (CAS), made significant progress in the intersection of artificial intelligence and neuroscience. The team proposed a brain-like network model based on endogenous complexity, which was published in the journal Nature Computational Science. This research achievement provides a new perspective for the development of artificial intelligence, aiming to build broader and more universal cognitive capabilities.
The co-corresponding authors of the study include researchers Li Guoqiji and Xu Bo from the Institute of Automation, CAS, and Professor Tian Yonghong from Peking University. The co-first authors are undergraduate students He Linxuan from the Qian Xuesen Class at Tsinghua University, Xu Yunhui from the Department of Mathematical Sciences at Tsinghua University, Ph.D. students He Weihua and Lin Yihan from the Department of Precision Instrument at Tsinghua University.
The current development of artificial intelligence is mainly dependent on methods based on extrinsic complexity, which involves building larger, deeper, and wider neural networks to improve model performance. However, this approach faces challenges such as high computational resources and energy consumption, as well as insufficient interpretability. To overcome these difficulties, the research team drew inspiration from the complex dynamics of brain neurons and proposed a method for constructing brain-like neural network models based on endogenous complexity.
The team’s concept of “small network models with endogenous complexity” involves simulating the complex dynamics of biological neurons to introduce complex internal structures into individual neurons, thus constructing more efficient AI models. The researchers used the HH model, which replaces the traditional LIF model in pulse neural networks. The HH model is a mathematical model that describes the mechanism of action potential generation in neurons, capable of simulating the response of neurons to various stimuli.
Through theoretical proofs and simulation studies, the research team demonstrated the equivalence of the HH model and the LIF model in the generation of action potentials, as well as how to enhance the endogenous complexity of computing units through the design of microstructures. This research provides a new method for constructing AI models, which is expected to play a significant role in future AI research.
This research achievement not only provides a new theoretical foundation and technical path for the development of artificial intelligence but also offers new perspectives and methods for neuroscience research. With the deepening collaboration between artificial intelligence and neuroscience, breakthroughs are expected to be made in the future in fields such as cognitive science and the treatment of neurological diseases.
【来源】https://www.jiqizhixin.com/articles/2024-08-18-2
Views: 4