Leveraging Recurrent Neural Networks for Predicting Motor Movements from Primate Motor Cortex Neural Recordings

TL;DR

This paper introduces an Autoencoder Gated Recurrent Unit model for decoding primate motor movements from neural recordings, achieving high accuracy and efficiency, and winning a neural decoding challenge.

Contribution

The paper presents a novel AEGRU architecture with pruning techniques that improve decoding performance and computational efficiency in neural movement prediction.

Findings

Achieved 0.71 R^2 score surpassing baselines

Ranked first in IEEE BioCAS 2024 Neural Decoding Challenge

Reduced MAC operations by 41.4% with minimal performance loss

Abstract

This paper presents an efficient deep learning solution for decoding motor movements from neural recordings in non-human primates. An Autoencoder Gated Recurrent Unit (AEGRU) model was adopted as the model architecture for this task. The autoencoder is only used during the training stage to achieve better generalization. Together with the preprocessing techniques, our model achieved 0.71 $R^2$ score, surpassing the baseline models in Neurobench and is ranked first for $R^2$ in the IEEE BioCAS 2024 Grand Challenge on Neural Decoding. Model pruning is also applied leading to a reduction of 41.4% of the multiply-accumulate (MAC) operations with little change in the $R^2$ score compared to the unpruned model.