基于受限玻尔兹曼机的神经元群体响应的贝叶斯解码

Bayesian Decoding of Neural Population Responses Based on the Restricted Boltzmann Machine

神经元通过尖峰模式传递有关刺激的信息,多个神经元通过突触相互联系,构成了复杂的神经回路。在过去的一个世纪中,多电极记录技术的进步使科学家们能够获取一个完整神经回路的细胞响应。这些记录表明,神经元的活动之间存在显著相关性。因此,本文提出利用受限玻尔兹曼机模型描述神经元响应活动之间的相关性,建立神经元群体响应的编码模型,并利用贝叶斯定理构建了基于受限玻尔兹曼机模型的解码器,将它应用于模拟的小鼠视觉皮层神经元的响应序列中。实验结果表明,此解码器在准确率方面优于不考虑神经元之间相关性的独立模型解码器。

Neurons represent the information about the stimuli by spiking pattern,and the neurons connect to each other through their synapses,forming complex neural circuits.Over the past century,the advances in multi-electrode recording technology have allowed scientists to get records from cells in a complete neural circuit.These records have shown that there is a significant correlation between the activities of neurons.Therefore,we use the Restricted Boltzmann Machine(RBM)to describe the correlation for the activity of neural population and establish an encoding model from the RBM.Then,the Bayesian theorem are used to construct a decoder based on the RBM,which been applied to the simulated neural responses of the mouse visual cortex neurons.The experimental results show that the proposed decoder has higher accuracy compared with the independent decoder.