The browser you are using is not supported by this website. All versions of Internet Explorer are no longer supported, either by us or Microsoft (read more here:

Please use a modern browser to fully experience our website, such as the newest versions of Edge, Chrome, Firefox or Safari etc.

Deep Generative Models

“Deep Fakes” for Good: Optimising the Scientific use of Deep Generative Models. ASG 2022

Deep Generative Models (DGMs ) — often referred to as deep fakes — have become famous because of their use to make fake videos of celebrities. However, artificial data produced using DGMs can be used to solve problems across a broad range of scientific disciplines. At a technical level DGMs use deep neural networks to produce synthetic data which has the same properties as real data.

The DGM ASG will bring together leading practitioners of this AI technique with researchers drawn from four faculties in order to explore the possible application of DGMs in a broad range of research disciplines. 

DGM ASG will begin by considering five science domains: astronomy, human geography, materials science, medicine and particle physics. The ASG therefore spans an enormous range of length scales: from the size of an individual atomic nucleus to the diameter of the Milky Way Galaxy. However, researchers face similar challenges in all five science areas: with the help of DGMs, we wish to de-noise, compress and search images, extract features and detect anomalies, as well as impute missing data and reduce dimensionality of our data. 

Deep Fakes lead to a number of important questions for society as a whole, concerning both the safety and legal issues as well as the ethics of artificial data. Therefore, DGM ASG will host two workshops during the spring 2023. Both workshops will include experts from LU and beyond to address a number of issues and questions, including: ownership of artificial data; how to identify and counteract misuse of artificial data; and how to ensure compliance with future legislation.