Speaker: Joshua Glaser (Postdoctoral Researcher at Paninski and Cunningham Labs, Columbia University)

Title: Machine learning for neural decoding

Abstract: Despite rapid advances in machine learning tools, the majority of neural decoding approaches still use traditional methods, like linear regression. Improving neural decoding algorithms allows us to better understand the information contained in a neural population, and can help advance engineering applications such as brain machine interfaces. In the first part of this talk, I'll describe work where we applied modern machine learning techniques to decode behavior from spiking activity across several brain areas. We found that modern methods, in particular neural networks and ensembles, significantly outperformed traditional approaches, even with limited data. We have created a code package to facilitate wider implementation of these methods. In the second part of the talk, I'll describe work in which we aimed to build robust decoders to predict muscle activity from motor cortex activity across multiple behavioral tasks. While a linear decoder performed well within a single task, we found that nonlinear decoders were necessary to accurately predict EMGs over multiple tasks. We demonstrated the ability of an LSTM to act as a robust decoder, and also created more interpretable models to better understand the nature of the nonlinearity. Across the two studies, our results demonstrate the importance of modern machine learning techniques for neural decoding.