Neural Decoding for Location of Macaque’s Moving Finger Using Generative Adversarial Networks

In this paper, we will study the problem that the location of a macaque's moving finger is decoded through neuron spike signals in its motor cortex. In traditional neural decoding methods, supervised learning is more popular because their decoding accuracy could be guaranteed. However, this paper proposes a weak supervised learning (WSL) method to decode the location of a macaque's moving finger and an unsupervised deep learning model, generative adversarial network (GAN) is introduced to optimize a training model and ultimately achieve an accurate estimation of the finger movement trajectory. In experiments, we use public data to evaluate the decoding performance of the proposed method. The experimental results show that this WSL method has higher accuracy of the location decoding.