Artificial and biological neural networks in cognitive tasks

Recent progress in both AI research and neuroscience open novel synergies for advancing both fields. Recent studies have shown that biologically realistic structures such as neural representations of navigation appear spontaneously during supervised learning of navigational tasks in artificial neural networks.

This opens an alternative avenue to study and gain insights into learning processes that can help understand the structure and dynamics in biological neural networks. These models do not include the level of physiological detail of computational neuroscience models, but instead focus on the emergent properties of learning networks.

The lack of knowledge about contributing elements to network processes limits the biological relevance of theoretical models. Moreover, models often lack dynamic processes such as plasticity.

A promising path is implementation of deep learning networks to compare with learning in the brain. However, these networks are usually simplistic with no contact to neural networks of the brain.

Deep insight into the network architecture, dissociation of neural contribution and behavioural outcomes hold great potential, though yet unexplored, to develop and explore processes in biologically inspired artificial neural networks. Moreover, large-scale data of high temporal and spatial resolution allow direct comparison with simulations performed in models.

The candidate will integrate experimental aspects such as cell-types and molecular features into artificial neural network models, extend the models to address reinforcement learning, apply the models to learning scenarios similar to that of the experiments, and compare network structure and dynamics during learning in biological and artificial systems using techniques developed for the experimental systems.

This two-pronged approach - combining experimental studies and exploratory modeling - will help us develop a new level of understanding of robust learning mechanisms in these systems and change the way experiments are conducted.

The project include 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 in statistical physics, scientific programming or implementation and studies of neural network systems is an advantage.

Supervisors

Professor Marianne Fyhn

Professor Anders Malthe-Sørenssen

Call 2: Project start autumn 2022

This project is in call 2, starting autumn 2022. 

Published Sep. 9, 2020 10:50 AM - Last modified Oct. 22, 2020 4:30 PM