A Novel Spike Sorting Method Based on Semi-supervised Learning

Abstract

Spike sorting is a prerequisite technique for neural coding and decoding research. So far many manual and automatic spike sorting approaches have been proposed, however this issue is still a challenge due to high time consumption of human or unreliability of machine. In this study, a semi-supervised spike sorting framework was proposed, in which three clustering or learning components, i.e. self-organizing map (SOM), manifold learning and affinity propagation clustering (APC) work cooperatively. The SOM serves as the technique for data reduction, as well as the visualization of cluster structure, which enabled the operator’s ”supervised” intervention. The manifold learning technique describes the intrinsic structure underlain by SOM grid, and yields the similarity matrix as output. The affinity propagation clustering takes as input the yielded similarity matrix, followed by the human’s guidance, and outputs the reasonable clustering result. The advantage of our framework is the efficient combination of traditional and novel approaches, wherein the semi-supervised intervention is embedded appropriately to guide the clustering process. The practical experiments’ results show that the proposed approach performed very well.

This work was supported by the grant of the Guide Project of Innovative Base of Chinese Academy of Sciences (CAS), No.KSCX1-YW-R-30, and the grant of Oversea Outstanding Scholars Fund of CAS, No.2005-1-18.