Modeling Uncertainty in Neural Decoding: A Density-based Approach to Connect Neural Activity to Behavior

Author

Tianxiao He

Date of Award

2023

First Advisor

Amanda Landi

Second Advisor

Anne O'Dwyer

Abstract

This study aims to explore the connection between neural activity and animal behavior and to investigate the neural decoding problem that reconstructs behavior signals from spike activity. After reviewing the literature on neural decoding, we recognized an inaccuracy and inefficiency in spike sorting, a preprocessing step that converts raw electric signal to neural activity of each neuron. To address this problem, we developed a density based decoding method that encodes the uncertainty of spike assignment using a Gaussian Mixture Model. The primary dataset comes from mice who have been placed in perceptual decision making experiments; we used a pipeline for preprocessing neural data and obtaining behavior of interest from video. Then we used a Gaussian Mixture Model to explore density-based neural representations, and applied this to decode the animals’ behavior. Our results show that for both perceptual choices and time-varying behavior, density based model outperforms thresholded method, and is comparable to spike sorted result.

Recommended Citation

He, Tianxiao, "Modeling Uncertainty in Neural Decoding: A Density-based Approach to Connect Neural Activity to Behavior" (2023). Senior Theses. 1641.
https://digitalcommons.bard.edu/sr-theses/1641