AF: Small: Collaborative Research: A Computational Theory of Brain Function

这个项目旨在识别、探索、严谨地呈现和验证“大脑是如何工作的”这个谜题的一个解决方案。这是所有科学领域中最基本、最具挑战性的前沿之一。大脑中的计算将在一个中等规模的水平上进行，远远大于单个神经元和突触，但明显小于整个大脑。核心假设是“集合”，即大量且高度互联的神经元集合，是大脑计算的引擎。在这个层面上研究计算对于理解高级认知功能至关重要，尤其是人类的认知功能，如推理、规划和语言；这种脑计算的表述特别适合于计算理论的方法论和观点，并将进一步扩大其影响范围。这个项目是典型的跨学科，将为研究生和本科生提供多方面的培训，包括计算机科学理论、机器学习、认知神经科学和心理学。它将在这个特殊的科学界面上开发跨学科的研究生课程。该项目的成果将通过所有这些学科的会议和期刊，以及座谈会和公开讲座广泛传播，而各种背景的学生将参与最前沿的研究体验。程序集可以是强大的计算系统的基础，该系统涉及包括项目、关联和合并在内的一系列操作。通过定理和模拟，这些操作可以被证明是可信的（也就是说，它们可以被“编译”到神经元和突触的水平）和有用的（在某种意义上，它们可以帮助解释现有的实验结果）。该项目将通过以下方式来实现这个装配假设：(1)扩展我们的建模和数学分析技术来研究装配计算；(2)开发更准确、更高效的仿真方法；(3)着手在形式计算之外的新模式下对装配的计算能力进行多管齐下的探索，特别是(a)通过模式完成的概率和动态系统类计算和(b)学习和预测；(4)数学建模和算法研究突触连接的动态和偏差，以及组装重叠，影响大脑计算的各种模式；更重要的是，(5)功能磁共振成像（fMRI）实验，并通过新的算法和机器学习技术分析这些实验结果和现有的皮质电图（ECoG）数据，以识别人脑中组装计算的证据。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 seeks to identify, explore, render rigorous, and validate one piece of the solution to the puzzle "how does the brain work?" - one of the truly fundamental and most challenging frontiers in all of science. Computation in the brain will be approached at an intermediate level of scale, far larger than that of individual neurons and synapses yet significantly smaller than that of the whole brain. The core hypothesis is that "assemblies," large and highly interconnected sets of neurons, are the engine of brain computation. Studying computation at this level is crucial for understanding higher cognitive functions, especially in humans, such as reasoning, planning, and language; and this formulation of brain computation is particularly amenable to the methodology and point of view of the theory of computation, and will further its reach. This project is quintessentially interdisciplinary, and will provide multi-faceted training to graduate and undergraduate students in Computer Science Theory, Machine Learning, and Cognitive Neuroscience and Psychology. It will develop interdisciplinary graduate courses in this particular scientific interface. The results of the project will be disseminated broadly via conferences and journals in all these disciplines, but also in colloquia and public lectures, while students of a great variety of backgrounds will participate in a cutting-edge research experience. Assemblies can be the basis of a powerful computational system involving a repertoire of operations including project, associate, and merge. These operations can be shown, through theorems and simulations, to be plausible (that is, they can be "compiled down" to the level of neurons and synapses) and useful (in the sense that they can help explain extant experimental results). The project will pursue this assembly hypothesis through: (1) expanding our modeling and our mathematical techniques of analysis for the study of assembly computation; (2) developing more accurate and efficient simulation methodology; (3) embarking on a multi-pronged exploration of the computational power of assemblies in novel modalities beyond formal computation, in particular (a) probabilistic and dynamical systems-like computation through pattern completion and (b) learning and prediction; (4) mathematical modeling and algorithmic investigation of the ways in which the dynamics and biases of synaptic connectivity, as well as assembly overlap, affect the various modes of brain computation; and, importantly, (5) functional magnetic resonance imaging (fMRI) experiments, and the analysis of the results of these experiments and extant electrocorticography (ECoG) data through novel algorithmic and machine learning techniques for the purpose of identifying evidence of assembly computation in the human brain.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.