一种胶质细胞偶联人工神经网络模型

Glia-coupled Artificial Neural Network

受到生物大脑神经胶质细胞和神经元互作用机制的启示,提出一种胶质细胞偶联人工神经网络(Glia-Coupled Artificial Neuron Networks,GCANNs)模型.对当前人工神经网络模型中的基本单元—神经元进行扩展,将每一个神经元连接一定数量的胶质细胞作为网络基本单元,通过神经元和胶质细胞的相互作用模型进行基本单元各元素的计算和更新,最后通过遗传算法对该模型的参数、拓扑结构进行优化.该网络模型能反应胶质细胞和神经元之间的双向通讯机制,更接近神经系统各元素之间的信息处理方式.实验表明,该模型能提升网络学习性能,在图像、文本等的分类任务上表现出比传统人工神经网络更高的准确率.此外,基于遗传算法的网络优化实验表明,为获得更高的适应度,大部分神经元都选择了与一定数量的胶质细胞“相连”,说明胶质细胞偶联的模型是经过了“自然进化、优胜劣汰”的更优模型.该模型能为构建更符合大脑工作机理的神经网络提供新的思路和方法,也能反向启发和推进神经科学研究.

Inspired by the interaction mechanism between glia and neurons in the biological brain,we proposed a glia-coupled artificial neuron networks(GCANNs)model.As the basic unit of the artificial neural network,the neurons are extended.Each neuron is connected with a certain number of glial cells as the basic unit of the network.The values of each element of the basic unit are calculated and updated through the interaction model between neurons and glia.Finally,the parameters and structure of the network are optimized by the evolutionary algorithm.The network model can reflect the two-way communication mechanism between glia and neuronsand is closer to the information processing mode between the elements of the nervous system.Experiments show that the proposed model can improve the learning performance of the network,especially in the field of classification for images,text,and so on.In addition,the network optimization experiments based on genetic algorithm show that most neurons have chosen to“connect”with a certain number of glial cells in order to obtain higher fitness,indicating that the glia-coupled artificial neuron network is a better model with“natural evolution and survival of the fittest”.The model provides new ideas and methods for constructing neural networks that are more consistent with the working mechanism of the brain,and can also inspire and promote the research of neuroscience in reverse.