基于CUDA的多信道锋电位实时分类方法

为提高多信道神经元锋电位分类任务的计算效率,满足其在实时场景下的应用需求,提出基于统一计算设备架构(compute unified device architecture,CUDA)的掩蔽高斯混合模型的并行化实现和优化方案。利用高维锋电位数据的稀疏特性和高斯混合模型的强抗干扰性以及良好并行性,借助GPU图形处理器,对特征掩蔽高斯混合模型(Masked Gaussian mixture model,Masked GMM)进行并行实现,进行针对性优化。实验结果表明,在32信道的锋电位数据集上,与原有的CPU串行实现相比,该方案分类速度提高了170倍左右,达到了实时计算,为高维信道锋电位实时分类提供了可行的解决方案。

Multi-channel real-time spike sorting method based on CUDA

To improve the computational efficiency of multi-channel spike sorting task and meet its application requirements in real-time scenarios,parallel implementation and optimization of masked Gaussian mixture model based on compute unified device architecture(CUDA)was proposed.The sparse characteristics of high-dimensional frontal potential data and the strong anti-interference of Gaussian mixture model and good parallelism were utilized.The GPU graphics processor was used to implement the Masked Gaussian mixture model(Masked GMM)in parallel.Targeted optimization was carried out.Experimental results show that the classification speed of the method is improved by 170 times compared with the original CPU serial implementation on the 32-channel spike dataset,and the real-time calculation is achieved,it provides a feasible solution for high-dimensional channel real-time spike sorting.