NeuroLearn

Inspiration

Throughout my four years as a research assistant of the Golshani Lab at the David Geffen School of Medicine, I have had a comprehensive involvement in the complete research process, collecting data firsthand and developing data analysis on MatLab.  In this project, mice were placed in an audio or visual task, where they had to discern between two distinct signals in order to obtain a reward. I observed instances where mice performance dropped unexpectedly, prompting us to slow down training and debug external factors that could account for these changes. I created this project to predict expected performance statistically in order to act as a diagnostic tool to determine whether drops in performance were statistically regular or were the result of underlying external issues (prompted by increases in days to switch). Additionally, this tool provides insight to whether the mouse is learning at an expected rate using past training cycles.

Function

This tool uses an XGBoost model trained on complete mice learning cycles conducted in the past to predict days until the mouse achieves learning mastery, where the training modality (audio vs. visual) can then be switched. 

Relevant Skills

XGBoost, SHAP, scikit-learn, SciPy, Matplotlib, MatLab, HTML