🔒 How it’s done in science (4)

Change exactly what I want

Marek Śmieja is researching new methods of controlling the data generation process in artificial intelligence models. His project focuses on developing deep conditional generative models that allow to precisely shape the outcomes created by algorithms like GANs and VAE. These models are used in many fields, from explaining AI decisions, through generating counterfactual examples, to creating chemical molecules with specific properties or realistically completing missing parts of images.

One of the key challenges in this area is the high computational complexity and the need to adjust models to new data, which often requires resources comparable to those available to global tech companies. In order to address that, Śmieja’s team is working on plugin networks that will be able to modify data based on existing models, thus reducing computational costs.