Learning to see without a teacher: Using deep learning techniques to discover how the brain develops visual perception

Samenvatting

The brain is the most intelligent information processor we know, but it does not come into the world this way. Most of the brain’s intelligent functions have to be learned from experience with the world. The key to understanding the brain, therefore, is to understand how the brain learns. I will target this important question, by combining knowledge about the brain and the environment from which it learns, with insights from self-learning computer algorithms. Thanks to recent, exponential developments in these algorithms, we are now in a position to apply similar techniques to model learning in the brain. Using visual perception as a test bed, I will adapt existing supervised learning methods into a new computational model of unsupervised learning in the brain’s visual cortex. From this model, I will distil concrete, testable predictions that I will validate against data from human participants performing perceptual tasks. By thus dovetailing computational and empirical methods, this research aims to understand how neurons wire together into complex information-processing networks. This not only addresses a fundamental and outstanding question in our understanding of the brain, but may also aid the development of more advanced self learning computer algorithms based on the same principles.

Kenmerken

Projectnummer

019.182SG.014

Hoofdaanvrager

Dr. R.S. van Bergen

Verbonden aan

Radboud Universiteit Nijmegen, Donders Institute, Cognitive Neuroimaging

Looptijd

01/03/2019 tot 28/02/2021