Learning How to Learn Using Computational Models of Motivation
Computational models of motivation – such as novelty, interest and curiosity – can provide a way for artificial agents to select their own learning goals. Agents identify highly motivating experiences then learn how to repeat those experiences. Existing work with motivated agents has focused agents with a fixed learning mechanism to learn about highly motivating goals. This project will develop new motivated learning approaches in which agents are also motivated to select between different action approaches in different situations. This may include reflexive responses, learning by trial-and-error or learning by mimicry.
Description of Work:
- Review of psychological literature describing motivation and learning.
- Design computational models of motivation that can mediate between different learning approaches.
- Implement the models in simulation and at least one live application, such as virtual world or the Lego Mindstorms NXT robotic platform.
- Evaluate the models using empirical metrics and/or case studies. This may include comparison to existing computational models of motivation or motivated learning.