Trustworthy iterative deep-learning models and … hot air balloons
by Dr. Nicola Pezzotti (Director of Data Science at Philips Cardiologs Paris/ Assistant professor at the Eindhoven University of Technology)
Iterative, or unrolled, deep-learning models are becoming a cornerstone of several applications. Such models build on the existing theory of iterative algorithms and substitute, in a form or another, traditional operators or functions with learned ones.
In this talk, I will present a layman intuition – using hot air balloons! – on what makes such approaches different from traditional models, and how particular operations, such as data-consistency, are key to ensure trustworthiness of the produced output.
I will use some of my work on MRI reconstruction, interpretable AI and uncertainty estimation to dive deeper in the concepts above. I will then conclude with some thoughts on how such approaches relate to generative AI.
We have two more speakers at this event:
- Geert Litjens (Radboud UMC) will tell us about AI model robustness and transparency in medical imaging.
- Nicola Strisciuglio (University of Twente) will address the problems of robustness and generalization of computer vision models, and link them to characteristics of and bias in the training data.
Share this Post