About the project
Neural networks form the bedrock for modern artificial intelligence and is an essential tool for modeling data in important domains such as computer vision and natural language processing. However, the fundamental understanding of why neural networks generalize so well is still underdeveloped, and major innovations are necessary to improve this understanding. Information theory provides a natural mathematical framework for improving this understanding, since the processing pipeline of neural networks can be interpreted as a communication channel that compresses data into a new representation. This project aims to develop novel information theoretic measures that will reveal new insights into the inner workings of neural networks, which can improve our understanding of neural networks as a whole.
Project team
-
Associate Professor Kristoffer Wickstrøm, UiT The Arctic University of Norway, Department of Physics and Technology
-
Associate Professor Ali Ramezani-Kebrya, University of Oslo, Department of Informatics
-
Professor Robert Jenssen, UiT The Arctic University of Norway, Department of Physics and Technology