“AI Outside the Comfort Zone”
Despite the impressive development of AI in the last decade, there are certain areas of progress where advancements have been limited. One example is the lack of model robustness when faced with test data that differs from the training data. In this presentation, I will discuss the challenges arising from this issue and try to justify why the dominant AI paradigm of scaling models will not help us find a fundamental solution to this problem.
Bio
Sebastian Cygert is a postdoc at Ideas NCBR and an assistant professor at the Gdańsk University of Technology, where he defended his PhD in 2022. Previously, he worked as an applied scientist at Amazon, contributing to projects such as the visual perception system for the autonomous Amazon Scout robot and speech synthesis for Alexa. In addition to this, he has extensive industry experience, including work on mathematical modeling for Moody’s Analytics and involvement in local startups.
His research focuses on developing reliable and trustworthy machine learning models that can robustly adapt to dynamic environments. His work has been published at leading AI conferences such as ECCV (three papers in 2024), NeurIPS, and ICLR and he is a member of the ELLIS. Additionally, he is interested in the medical applications of AI, particularly through collaboration with the Gdańsk Medical University on a project aimed at early cancer diagnosis using liquid biopsies.