We are pleased to announce that Daniel Cremers, Professor at the Technical University of Munich and a leading figure in computer vision and machine learning, will deliver the upcoming talk in the ELLIS Webinar Series From Molecules to Models, part of the AI for Good Initiatives.

Convolutional networks have transformed image analysis and data science, becoming one of the driving forces behind the current AI revolution. In this session, the speaker will outline key developments in the field, beginning with the foundational breakthrough in large-scale image classification (Krizhevsky et al., NeurIPS 2012).

The talk will then explore how the capabilities of deep networks have been extended far beyond image classification, addressing new challenges in image analysis, shape analysis, medicine, and the life sciences. Illustrative applications include:

• Optical flow estimation (Dosovitskiy et al., ICCV 2015)
• Acceleration of diffusion tensor imaging (Golkov et al., 2016)
• Tuberculosis screening from chest X-ray images (Golkov et al., Scientific Reports 2019)
• Protein structure prediction (Golkov et al., NeurIPS 2016)

The session will conclude with an introduction to convolutional networks for graph-valued data, featuring a recent method that extends convolutional networks to directed graphs through holomorphic functional calculus (Koke & Cremers, ICLR 2024).

Learning Objectives

Explain the fundamental principles and architecture of convolutional networks

Apply convolutional networks to non-Euclidean data structures

More information here