Data Science Seminar

Uncertainty, causality and generalization: Attempts to improve image-based predictive modelling

Click to watch the video

This talk will give an overview of some recent work by our team on various aspects of predictive modelling in imaging. We will discuss how the language of causality can shed some new light on key challenges in machine learning for medical imaging, namely data scarcity and mismatch. For the latter, we present a meta-learning algorithm for domain generalization. We also look at some very recent results of our attempt to generate counterfactual images using deep structural causal models. Finally, we introduce a simple yet effective component for modelling spatially correlated aleatoric uncertainty in image segmentation which can be plugged into any existing network architecture. The resulting stochastic segmentation networks predict multiple plausible segmentation maps with a single forward pass.

Biosketch Ben Glocker

Ben Glocker is Reader (eq. Associate Professor) in Machine Learning for Imaging at Imperial College London. He holds a PhD from TU Munich and was a post-doc at Microsoft and a Research Fellow at the University of Cambridge. His research is at the intersection of medical image analysis and artificial intelligence aiming to build computational tools for improving image-based detection and diagnosis of disease.

 

Joint event with heidelberg.ai

to top
powered by webEdition CMS