Deep Learning: From Alchemy to Science

Deep Learning: from Alchemy to Science

Project Team: Ben Adcock (Mathematics, SFU), Simone Brugiapaglia (Mathematics, SFU), Nick Dexter (Mathematics, SFU). 

This project addresses fundamental questions concerning Deep learning (DL) and its use in practice. DL has caused a revolution in artificial intelligence and has rapidly penetrated diverse areas of everyday life. It has already led to a wave of automation, replacing human expertise by algorithmic decision-making. However, DL, as it is currently used, is highly unstable. Imperceptible changes in the input (e.g. an image) can cause a completely incorrect outcome (e.g. its classification). This issue has severe consequences for security, privacy, and in general, reliability of deep learning-based tasks. These issues can only be addressed by tackling the mathematical foundations of deep learning. This project proposes to do this by developing a framework for robust, accurate and efficient DL.