Sekeh Lab

This research project focuses on investigating the problem of Continual Learning. Continual Learning handles the training of multiple tasks sequentially, ideally both compressing the information from each task into a single architecture and leveraging knowledge from previously learned tasks to better perform each subsequent task. We aim to utilize the concept of Information Flow, a measure of how well a given connection passes information through the network, to better determine which pathways in the network are important for each given task. This investigation seeks to lay a theoretical basis for the use of Information Flow in improving Continual Learning methods, alongside the necessary empirical experiments to support its application.