RI: Small: Theory of Robust Learning Based on the Structure and Function of the Cortical Column

大脑如何学习，并在此过程中改变其突触连接，仍然是现代科学最大的谜团之一。该项目的目的是揭示强大的联想学习和长期记忆存储对突触连通性的影响，从而为这些基本大脑功能的定量分析奠定基础。研究者建议通过皮质回路建立一个健壮的联想学习的生物学现实模型。该模型将从一个单一的假设中推导出来，根据该假设，在给定的成人皮层回路中，突触连接处于稳定状态。在这种状态下，脑回路的联想记忆存储容量最大，学习新的联想的同时也会忘记一些旧的联想。该模型将整合目前关于兴奋性和抑制性神经元类别的知识，以及皮层神经元的轴突和树突乔木形态所施加的结构连接约束，以及突触连接数量和强度的稳态约束。本文提出了一种模拟稳态学习的方法，该方法是基于哺乳动物新皮层中研究得最好的网络之一——啮齿动物体感皮层的桶中心柱。模拟将嵌入柱的结构连通性，从不同皮层深度的三维重建的神经元形态中建立。稳态电路的显著特征将通过大型实验研究数据集进行验证，这些实验研究报告了神经元之间连接的概率、特定高阶连接基序的概率、统一突触后电位的分布以及啮齿动物桶状皮质层间和层间投影的相对强度。该数据集将作为项目的一部分创建，并将包括所有皮层中存在的主要兴奋性和抑制性细胞类别的连通性。拟议的研究植根于统计学习的基本原理，并将通过基础神经科学和计算智能应用，推进认知功能的理论和计算建模的最新状态。

How the brain learns, and in the process modifies its synaptic connectivity, remains one of the greatest mysteries of modern science. The objective of this project is to uncover the effects of robust associative learning and long-term memory storage on synaptic connectivity, thus creating the basis for quantitative analyses of these fundamental brain functions. The investigator proposes to develop a biologically realistic model of robust associative learning by cortical circuits. The model will be derived from a single hypothesis, according to which synaptic connectivity in a given circuit of adult cortex is functioning in a steady-state. In such a state the associative memory storage capacity of the circuit is maximal, and learning new associations is accompanied with forgetting some of the old ones. The model will integrate current knowledge of excitatory and inhibitory neuron classes, with structural connectivity constraints imposed by the morphologies of axonal and dendritic arbors of cortical neurons, with homeostatic constraints on numbers and strengths of synaptic connections. It is proposed to simulate steady-state learning based on one of the best studied networks in the mammalian neocortex - the barrel-centered column of rodent somatosensory cortex. The simulations will be imbedded in the structural connectivity of the column, built from the morphologies of neurons reconstructed in three-dimensions from various cortical depths. Salient features of steady-state circuits will be validated against a large dataset of experimental studies reporting probabilities of connections between neurons, probabilities of specific higher-order connectivity motifs, distributions of unitary postsynaptic potentials, as well as relative strengths of laminar and inter-laminar projections in rodent barrel cortex. The dataset will be created as part of the project and will encompass connectivity of major excitatory and inhibitory cell classes present in all cortical layers. The proposed research is rooted in the basic principles of statistical learning and will advance the state of the art in theoretical and computational modeling of cognitive functions with basic neuroscience and computational intelligence applications.