Foundations for learning transferable concepts
Transfer learning (TL) studies the automatic acquisition and application of transferable knowl- edge (Pan and Yang, 2009). In recent years, a large body of (primarily empirical) research has studied TL in the context of prediction, using algorithms built on surprisingly shaky mathemat- ical foundations (as discussed in Johansson et al. (2019); Zhao et al. (2019)). While there are known conditions that guarantee successful TL, these are often violated in fundamental ways in practice. This project aims to address this gap by providing rigorous, plausible and sufficient conditions for TL and practical learning algorithms that make use of them as assumptions. The hired student will pursue a PhD in machine learning within computer science and engineering.
This page is only available in english
- Wallenberg AI, Autonomous Systems and Software Program (Non Profit, Sweden)