大脑皮层网络连接的计算机仿真

Computer Simulation of Connectivity in Cerebral Cortex Neural Networks

为了研究动物大脑皮层网络连接特征 ,了解大脑在特定环境下的演化适应性 ,为机器学习提供新的思路 ,该文用图表示网络结构 ,用匹配复杂度表示外界刺激下网络的动力学特性 ,采用图选择的方法对随机图进行变异和选择 ,确定系统连接与感觉层间的匹配关系。仿真结果表明 :由图选择获得的网络结构 ,呈现若干密集的神经元群 ,神经元群间松散连接 ,特定的结构模式可以使系统与感觉层神经元统计结构间达到最大程度的匹配 。

To simulate the connectivity of higher vertebrates' cerebral cortex and its specific evolutionary adaptations, which could provide a new way of machine learning, the paper represents the networks structures with graphs and describes their dynamics with matching complexity as they respond to extrinsic stimulus. Then it selects and mutates random graphs using graph selection method to determine how well the intrinsic correlations match the statistical structure of the system sensory input. The simulation results show that: the graphs decided by graph selection present dense neuron groups linked by a relatively small number of reciprocal bridges. It also shows that special structure could fit the sensory statistical structure most. The results have the similar features to the real cerebral cortex.