Collaborative Research: MEMONET: Understanding memory in neuronal networks through a brain-inspired spin-based artificial intelligence

合作研究：MEMONET：通过受大脑启发的基于自旋的人工智能了解神经元网络中的记忆

• 批准号: 1939992
• 负责人: Ying Zhang
• 金额: $ 39.96万
• 依托单位: University of Rhode Island
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2022-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1939992&HistoricalAwards=false
• 关键词: Collaborative; Research; MEMONET; Understanding; memory

The brain is arguably the most sophisticated and the most efficient computational machine in the universe. The human brain, for example, comprises about 100 billion neurons that form an interconnected circuit with well over 100 trillion connections. Understanding how a multitude of brain functions emerge from the underlying neuronal circuit will give insights into the operating principles of the brain. In this award, a multidisciplinary team of systems biologist, computational biologist, material scientist, neuroscientist, and machine learning expert will work synergistically to leverage the data revolution in neuroscience to answer a fundamental question: How does the brain learn, store, and process information? The team will develop and apply advanced data analysis algorithms to harness the great volume of neuronal data generated by the latest imaging and molecular profiling technologies, for elucidating the neuronal circuits driving brain functions. Computer simulations of a spin-electronic (spintronic) device will further serve as a platform to validate and emulate important operational characteristics of such neuronal circuits. The award sets the groundwork for an interdisciplinary data science research and educational program that will bring a new and powerful paradigm for studying brain functions as well as for designing transformative brain-inspired devices for information processing, data storage, computing, and decision making.The project has a specific focus on an essential function of the brain: motor-skill learning. This function emerges from the underlying circuitry of neurons that governs the activities of molecular signal transmission and neuronal firing. Importantly, the neuronal circuit in a mammalian brain is highly plastic and dynamic, features that endow animals with the ability to respond to myriad external stimulations through learning. By harnessing the latest data revolution in neuronal imaging, single neuron molecular profiling, spintronic device simulation, network inference, and machine learning, a team of multidisciplinary investigators will be supported by this award to investigate the fundamental principle of neuronal circuit rewiring that drives brain?s learning function. More specifically, the team sets out to achieve the following specific tasks: (A) Infer learning-induced rewiring of large-scale neuronal networks from two-photon calcium imaging data through the development of novel and powerful network inference algorithms; (B) Build biochemical-based models of neuronal circuits by integrating molecular profiling with neuron firing and connectome dynamics; and (C) Develop a spintronic material network model that emulates learning and memory formation by exploiting the spin dynamics in spintronic materials. The project seeks to lay the foundation for the creation of an interdisciplinary data-intensive brain-to-materials initiative that will be applied to understand and emulate the operational principles of brain neuronal circuits underlying learning, cognition, memory formation, and other behaviors. The outcomes of the initiative will have a paramount impact on the society, not only in our understanding of the brain and its functions, but also in overcoming current bottlenecks of existing computing architectures. This project is part of the National Science Foundation's Harnessing the Data Revolution (HDR) Big Idea activity.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

大脑可以说是宇宙中最复杂和最有效的计算机器。例如，人类大脑由大约1000亿个神经元组成，这些神经元形成了一个具有超过100万亿个连接的互连电路。了解大量的大脑功能是如何从潜在的神经元回路中产生的，将有助于深入了解大脑的工作原理。在这个奖项中，一个由系统生物学家、计算生物学家、材料科学家、神经科学家和机器学习专家组成的多学科团队将协同工作，利用神经科学中的数据革命来回答一个基本问题：大脑如何学习、存储和处理信息？ 该团队将开发和应用先进的数据分析算法，利用最新的成像和分子分析技术产生的大量神经元数据，阐明驱动大脑功能的神经元回路。自旋电子器件的计算机模拟将进一步作为验证和仿真这种神经元电路的重要操作特性的平台。该奖项为跨学科的数据科学研究和教育计划奠定了基础，该计划将为研究大脑功能以及设计用于信息处理、数据存储、计算和决策的变革性大脑启发设备带来新的强大范式。该项目特别关注大脑的基本功能：运动技能学习。这种功能出现在神经元的底层电路中，该电路控制分子信号传输和神经元放电的活动。重要的是，哺乳动物大脑中的神经元回路具有高度的可塑性和动态性，这些特征赋予动物通过学习对无数外部刺激做出反应的能力。通过利用神经元成像，单神经元分子分析，自旋电子器件模拟，网络推理和机器学习的最新数据革命，一个多学科研究人员团队将获得该奖项的支持，以研究驱动大脑的神经元电路重新布线的基本原理。的学习功能。更具体地说，该团队着手实现以下具体任务：（A）通过开发新颖而强大的网络推理算法，从双光子钙成像数据中推断学习诱导的大规模神经元网络重新布线;（B）通过整合分子图谱与神经元放电和连接体动力学，构建基于生物化学的神经元回路模型;以及（C）开发自旋电子材料网络模型，其通过利用自旋电子材料中的自旋动力学来模拟学习和记忆形成。该项目旨在为创建跨学科的数据密集型大脑到材料倡议奠定基础，该倡议将用于理解和模拟学习，认知，记忆形成和其他行为的大脑神经元回路的操作原理。该计划的成果将对社会产生重大影响，不仅是在我们对大脑及其功能的理解方面，而且在克服现有计算架构的瓶颈方面。 该项目是美国国家科学基金会利用数据革命（HDR）大创意活动的一部分。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Divergent Learning-Related Transcriptional States of Cortical Glutamatergic Neurons.

皮质谷氨酸神经元的发散学习相关转录状态。

• DOI: 10.1523/jneurosci.0302-23.2023
• 发表时间: 2024
• 期刊: The Journal of neuroscience : the official journal of the Society for Neuroscience
• 作者: Dunton,KatieL;Hedrick,NathanG;Meamardoost,Saber;Ren,Chi;Howe,JamesR;Wang,Jing;Root,CoryM;Gunawan,Rudiyanto;Komiyama,Takaki;Zhang,Ying;Hwang,EunJung
• 通讯作者: Hwang,EunJung

Deep learning for neural decoding in motor cortex

• DOI: 10.1088/1741-2552/ac8fb5
• 发表时间: 2022-10-01
• 期刊: JOURNAL OF NEURAL ENGINEERING
• 影响因子: 4
• 作者: Liu, Fangyu;Meamardoost, Saber;Wang, Linbing
• 通讯作者: Wang, Linbing