URoL:EN: Learning the Rules of Neuronal Learning

该项目将探索神经网络在工程微环境中的生命规律。生物大脑中的神经元网络通常比人工计算机更有效、性能更好地解决模式识别和分类方面的挑战性问题。这启发了当今复杂的语音识别和计算机视觉背后的算法和计算工程模型，然而，在理解活神经元及其相邻神经元的物理机制方面，进展有限。该项目汇集了神经元的模式、记录、刺激和遗传操作方面的最新技术进展，研究如何培养健康的神经元，同时以足够精细的规模持续观察和刺激它们，以揭示单个神经元的各个部分如何对网络的整体学习和计算做出贡献。目标是在培养皿中通过神经元进行可编程计算的概念验证演示。如果成功，这将对自然神经元计算的科学理解产生重大影响，并将引入一套全新的工程工具，用于与活神经元相互作用，并探索计算上的可能性。对生命神经元规则的新认识可能会影响人工智能、机器人和神经假肢。该项目将寻求与公众接触新的见解和结果，同时提供软件和数据集作为科学资源，允许其他研究人员验证和扩展结果。神经元计算本质上是分布式的和动态的。该项目将通过扩展现有技术，在工程微环境中以前所未有的空间和时间尺度研究神经网络的生命规则，以证明培养神经元的小型网络具有可训练的时空模式识别。为了实现这一目标，一个由工程师和生物学家组成的多学科团队将采用新兴技术在有图案的微环境中培养神经元，并使用高密度微电极阵列和光遗传学记录/刺激神经元。主要的科学假设是，当一个神经元在其树突树上呈现输入尖峰的时空模式时，可以训练它对许多干扰物刺激中的一种模式做出高度选择性的反应。这项研究将沿着五个不同的方向组织：1)发展一个稳定的、不破裂的、神经元/胶质共培养，它接受持续的背景刺激，以避免感觉剥夺；2)识别和定位细胞之间的网络突触连接，以及基于峰值时间依赖的可塑性对突触强度的修改；3)树突树中观察到的计算基元的表征；在模拟中对时空模式的选择性学习，以及5)训练一个活的、培养的神经元对目标时空模式做出选择性反应。该实验将利用高密度微电极阵列和高分辨率荧光显微镜，该显微镜具有投射图案光的能力，用于光遗传刺激细胞。这项研究将建立对神经元如何单独和集体响应其动态环境的基本和实用的理解，开发响应的预测模型，并开发控制律来调整网络以适应特定的时空模式识别任务。该奖项反映了美国国家科学基金会的法定使命，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。

This project will explore the rules of life of neuronal networks in an engineered microenvironment. The networks of neurons in biological brains routinely solve challenging problems in pattern recognition and classification more efficiently and with better performance than man-made computers. This has inspired algorithms and engineering models of computing behind the sophisticated voice recognition and computer vision available today, yet there has been limited progress towards understanding the physical mechanisms within a living neuron and its neighbors underlying its remarkable capabilities. This project brings together recent technological advances in patterning, recording, stimulation, and genetic manipulation of neurons to study how to nurture a healthy culture of neurons while continuously observing and stimulating them at a scale fine enough to uncover how the individual parts of a single neuron contribute to the overall learning and computation of the network. The goal is a proof-of-concept demonstration of a programmable computation by neurons in a dish. If successful, this will have significant implications for both the scientific understanding of natural neuronal computation and would introduce a completely new set of engineering tools for interacting with living neurons and exploring what is computationally possible. New understanding from the neuronal rules of life may impact artificial intelligence, robotics, and neural prosthesis. This project will seek to engage the public with new insights and results while providing the software and datasets as a scientific resource to allow other investigators to verify and extend the results. Neuronal computation is distributed and dynamic in nature. This project will investigate the rules of life of neuronal networks at an unprecedented spatial and temporal scale within an engineered microenvironment by extending existing technologies to demonstrate trainable spatiotemporal pattern recognition by small networks of cultured neurons. To accomplish this, a multidisciplinary team of engineers and biologists will adapt emerging techniques for culturing neurons in patterned microenvironments and for recording/stimulating neurons using high density microelectrode arrays and optogenetics. The primary scientific hypothesis is that a single neuron, when presented with a spatiotemporal pattern of input spikes distributed on its dendritic tree, can be trained to respond highly selectively to one pattern out of many distractor stimuli. This research will be organized along five distinct thrusts: 1) development of a stable, non-bursting, neuron/glia co-culture that receives persistent background stimulation to avoid sensory deprivation, 2) identification and localization of network synaptic connectivity between cells and the modification of synaptic strength based on spike-timing-dependent plasticity, 3) characterization of computational primitives observed in the dendritic tree, 4) development of training protocols for single-neuron, selective learning of spatiotemporal patterns in simulation, and 5) training of a live, cultured neuron to selectively respond to a target spatiotemporal pattern. The experiments will make use of high-density microelectrode arrays and a high-resolution fluorescence microscope with the capability to project patterned light for optogenetic stimulation of cells. This research will establish fundamental and practical understanding of how the neurons individually and collectively respond to their dynamic environment, develop predictive models of the response, and develop control laws to tune the network for specific spatiotemporal pattern recognition tasks.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.