Incremental Learning Agents and Human Interaction

Project: Dissertation

Project Details


[Draft scope]
The project is starting in computer vision and image analysis to become familiar with current methods and a well established domain. This is used as an arena to find a machine task that involves human interaction and a non-static environment that allows incremental learning. Designing the task and learning process to benefit from transparency of the agent decisions (explainability) is a secondary goal.

While deep learning with large amounts of up-front learning data has made recent success, an agent that is able to learn incrementally from data available during usage should be more versatile. A human-in-the-loop approach, where the user is taking an active part in the learning or acting process, could be one way to give transparency but the human should not become a bottle-neck with no added value.
Effective start/end date2019/01/01 → …