Scientific seminars by Dr. Zheng Zhao and Dr. Jennifer Alvén

We have two candidates for an assistant professorship at the Computer Vision Group: Dr. Zheng Zhao and Dr. Jennifer Alvén.

They will give two scientific seminars on Tuesday 28 November about their research. Please come and listen to them!

Time:
- Kl 14.00 Zheng Zhao, Uppsala University
- Kl 15.00 Jennifer Alvén, Chalmers

Title: Supervised generative diffusions and Schrödinger bridges
Speaker: Zheng Zhao

Abstract: Recently, diffusion models have been widely applied to generative modellings, and they are empirically shown to be the state of the art. The core of diffusion models lies in the stochastic differential equations (SDEs) that bridge two distributions, and the techniques (e.g., score matching and Schrödinger bridges) to find and simulate these SDEs. In this talk, we show the classical score matching approach for generative modelling, and then we show how to generalise and speed-up the problems via Schrödinger bridges. More importantly, we extend the two methods to solve a supervised generative modelling problem that additional exploits the pairing information in the training data. We demonstrate the methodology on image restoration tasks (e.g., super-resolution and denoising). This talk is related to our recent publication in ICML 2023 which can be at https://github.com/Algolzw/image-restoration-sde.


Title: From scanner to cluster - Usable, reliable & generalizable deep learning for medical image analysis
Speaker: Jennifer Alvén

Abstract: Automated tools with the capacity to make fast, accurate, and robust interpretation and analysis of medical images are in high demand within both medical research and clinical settings. My primary research objective is to develop medical image understanding methods empowered by state-of-the-art computer vision and machine learning techniques, placing particular emphasis on generalizability, reliability, and usability. In this seminar, I will provide an overview of my research endeavors as a postdoctoral researcher at the Sahlgrenska Academy and at Chalmers, where I've been involved in everything from designing data annotation protocols to implementing prototype models in hospital picture archiving and communication systems. The presentation will include applications such as coronary artery and plaque segmentation in cardiac CTA, ejection fraction prediction in echocardiography, outlier detection in cardiac CT and lymphoma classification in PET imaging, and topics such as explainable AI, uncertainty quantification, and semi-supervised learning.

Welcome!
Fredrik Kahl