While AI has made tremendous progress, in particular within deep learning, the brain is still far superior to machines in many areas: It is more energy efficient, it needs fewer examples to learn complex tasks, and it easily transfers strategies from one task to another through transfer learning. A particular challenge, which seems inherent to AI systems, is catastrophic forgetting in the process of continual learning, in which learning new tasks with the same AI machinery destroys previously learned knowledge.
In this project, you will develop bio-inspired and physics-based models, for example based on Hopfield networks, that combine local learning rules such as Hebbian learning with neural networks and apply the models to study continual learning and transfer learning tasks. The goal is to develop models that overcome catastrophic forgetting. You will learn to apply tools from physics, neuroscience and artificial intelligence to understand the mechanisms of learning and the formations of structures of representation.
These studies will be done in close collaboration with our leading experimental neuroscience lab which provides you with unique access to experimental studies of similar processes. We believe this close coupling between experiment and modeling and between physics, computer science and bioscience will be essential for next breakthroughs in our understanding of both artificial and biological neural networks.
The project includes collaborations with leading groups at Harvard University and the University of California San Diego.
Requirements
- You must have a master degree in physics, computational neuroscience or artificial intelligence.
- Documented experience from computational modeling, scientific programming or implementation and studies of neural network systems is an advantage.
Supervisors
Associate Professor Kai Olav Ellefsen
Call 2: Project start autumn 2022
This project is in call 2, starting autumn 2022.