The goal of this project is to interpret and understand the hidden units of a Deep Learning model trained on an Action Recognition task, particularly in classifying UCF-101 Action datasets.