Promotion lecture for Professor by Lars Hammarstrand

Promotion lecture for Professor by Lars Hammarstrand

Title: Learning Beyond Labels: Leveraging Unlabelled Data for Robust Perception

Abstract: The ability to reliably perceive and understand the environment is crucial for many real-world applications, including autonomous systems and decision-making in uncertain conditions. However, obtaining large-scale labeled datasets for training robust perception models is costly and often impractical. This talk explores how self-supervised and semi-supervised learning methods can leverage cheap, unlabelled data to build robust feature representations while reducing dependence on manual annotations.

We will discuss fundamental approaches that enable learning from unlabelled data and highlight applications in image classification, image segmentation, and autonomous driving. Additionally, we will examine challenges such as robustness to domain shifts and learning from out-of-distribution data, which are critical for real-world deployment. The talk concludes with key open research questions and future directions in the field.

Welcome!

Ragne Emardson and Lars Hammarstrand