EFRI-COPN: Dynamics of Neural Networks on a Planar Patch-Clamp Array: Training, Identification, and Control

EFRI-COPN：平面膜片钳阵列上的神经网络动力学：训练、识别和控制

• 批准号: 0835947
• 负责人: Russell Tedrake
• 金额: $ 188.16万
• 依托单位: Massachusetts Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-09-01 至 2012-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0835947&HistoricalAwards=false
• 关键词: EFRI; COPN; Dynamics; Neural; Networks

The core of this project will be a collaboration between Sebastian Seung, who has been using in vitro cultures to test his existing models of learning in neurons, and Tedrake, who has been working on intelligent robotics informed by control theory. The main ideas are: (1) to revolutionize the in vitro work, by solving technical problems with patch-clamp arrays, so that they can be used to monitor hundreds of cells effectively in parallel; and (2) to apply robust system identification to develop new reduced models of the living neural circuits (as well as calibrate traditional biological models), and train these circuits to address benchmark challenges which represent the cutting edge of research in robotics. The patch clamp arrays and the data which they generate may themselves be highly transformative. Solving these technical problems is a major step towards being able to interface with hundreds of neurons in the brain itself, in the future. The connection between control theory approaches, reinforcement learning and actual biological data, in facing common robotic control challenges, will encourage greater crossdisciplinary cooperation and understanding at an institution which plans a key role in US engineering education. The level of detail of this data will make it possible to evaluate models of learning in far more detail than is possible when data are available only on inputs and outputs of the neural circuits.

该项目的核心将是Sebastian Seung和Tedrake之间的合作，Sebastian Seung一直在使用体外培养来测试他现有的神经元学习模型，而Tedrake一直在研究基于控制理论的智能机器人。主要思想是：(1)通过解决膜片钳阵列的技术问题，彻底改变体外工作，使其可以有效地并行监测数百个细胞；(2)应用鲁棒系统识别来开发新的神经回路简化模型（以及校准传统的生物模型），并训练这些电路来解决代表机器人研究前沿的基准挑战。膜片钳阵列及其产生的数据本身可能具有高度的变革性。解决这些技术问题是未来能够与大脑本身的数百个神经元交互的重要一步。控制理论方法、强化学习和实际生物数据之间的联系，在面对常见的机器人控制挑战时，将鼓励一个计划在美国工程教育中发挥关键作用的机构进行更多的跨学科合作和理解。这些数据的详细程度将使评估学习模型的细节成为可能，而不是仅从神经回路的输入和输出中获得数据。