Lunch at ICAI: Computer Vision in NL

This Lunch at ICAI session is focused on computer vision in The Netherlands. Two ICAI labs share their story. Two speakers highlight their recent work in computer vision. And two leaders in the field point out future directions.

12.00 (noon): Bram van Ginneken (Radboud U. Nijmegen) presents the Thira Lab
12.05: Cristina González Gonzalo on Explainable AI in Medical Imaging
12.20: Cees Snoek (U. Amsterdam) presents the AIM Lab
12.25: Yingjun Du on Meta-Learning Batch Normalization
12.40: Bram van Ginneken and Cees Snoek discuss what's next in computer vision, in The Netherlands and beyond
13.00: End

All times are CEST.

Speaker: Cristina González Gonzalo, Thira Lab, Radboud UMC
Title: Explainable AI in Medical Imaging
Abstract: Deep learning (DL) systems in medical imaging have shown to provide high-performing approaches for diverse tasks in healthcare. Nevertheless, DL systems are often referred to as “black boxes” due to the lack of interpretability of their predictions. This is specifically problematic in healthcare applications, where it hinders experts’ trust and the integration of these systems in clinical settings. In this presentation, we highlight the importance of explainable artificial intelligence in healthcare and present our novel deep visualization method to generate interpretability of DL classification tasks in medical imaging, which we recently published in IEEE Transactions on Medical Imaging. The proposed method iteratively unveils abnormalities in a weakly-supervised manner and yields augmented visual evidence of the system’s predictions, including less discriminative areas that should also be considered for the final diagnosis. We show that augmented visual evidence highlights the biomarkers considered by clinical experts for diagnosis, improves the final performance for weakly-supervised lesion localization, and can be integrated in different interpretability frameworks.

Speaker: Yingjun Du, AIM Lab, U. Amsterdam
Title: Meta-Learning Batch Normalization
Abstract: While deep learning systems have provided breakthroughs in several computer vision challenges, they are still limited by their dependency on the availability of training data. Contrary to conventional approaches to Artificial Intelligence where a given task is solved from scratch using a fixed learning algorithm, meta-learning aims to improve the learning algorithm itself, given the experience of multiple learning episodes. Meta-learning provides an opportunity to tackle many of the conventional challenges of deep learning, including data and computation bottlenecks, as well as the fundamental issue of generalization. In this presentation, we mainly describe the contemporary meta-learning landscape, it includes the definition of meta-learning and different application scenarios (few-shot learning, domain generalization, few-shot domain generalization, etc). Finally, we will talk about our recent work: MetaNorm, a simple yet effective meta-learning normalization. MetaNorm tackles the challenging scenarios where the batch size is too small to produce sufficient statistics or when training statistics are not directly applicable to test data due to a domain shift. MetaNorm learns to learn adaptive statistics that are specific to tasks or domains. It is generic and model-agnostic, which enables it to be used with various meta-learning algorithms for different applications.

*** This is an online meeting. Make sure to (1) sign up for the meetup on the meetup page and (2) ensure you receive emails from Meetup. Shortly before the event we will send you the Zoom link and password to attend, as well as the info you need to log in via a web browser (if your organization does not allow you to install Zoom). You will only receive this if you have done both these steps. ***