Unsupervised neuron spike decoding for macaque's finger position via EM algorithm

Decoding for a macaque's moving finger position from neural signals in its motor cortex, is a problem about neural decoding. In the problem, many conventional decoding algorithms adopt a supervised training way, which needs some ground truth data and find a function w.r.t neural spikes and the finger's moving position. Therefore, the ground truth data will have an important effect on the performance of the conventional algorithms. This paper presents an unsupervised decoding algorithm using expectation maximization (EM) method. The advantage of the proposed algorithm is to be trained without ground truth data. Some experimental figures indicate that the proposed algorithm has fewer errors than the conventional algorithms under the condition of a small amount of supervised data adopted.