Virtual Talk with Martin Mundt
Abstract
Deep machine learning continues to advance the frontiers of innovative applications, recently through the rise of generative AI and foundation models. However, as much as their success is astonishing, their data-hungry and compute-heavy nature continues to entail an equal amount of scientific and socio-economic concerns. Repeated large-scale training is known to be unsustainable, the produced outputs are often misleading or even downright hallucinated, and the data-dependent development cycles shift privilege to those who have access. Inspired by the human’s remarkable ability to adapt efficiently without amassing tremendous amounts of data, lifelong learning promises to transcend these predicaments. In this talk, I will first walk through the elements of lifelong learning and how it provides remedies to successfully handle new situations, absolve us from the need to store large datasets, and equip models with efficient adaptation abilities. In turn, I will highlight how the resulting paradigm shift provides both a sustainable alternative to present large-scale trends and how the inherent ability to adapt efficiently over time is one imperative factor in enabling participatory AI that is developed for and with humans in mind.
Bio
Martin Mundt leads the Open World Lifelong Learning lab as an independent research group leader at the Hessian Center for Artificial Intelligence (hessian.AI) and interim professor (Vertretungsprofessor) at the Technical University of Darmstadt (TU Darmstadt). He is also a board member of directors at the non-profit organization ContinualAI, organizer at Queer in AI, and currently serves as diversity and inclusion chair for AAAI-2024 and review process chair for CoLLAs 2024 (Conference on Lifelong Learning Agents). Previously, he has received an M.Sc. in physics (2015) and obtained a PhD in computer science (2021) from Goethe University with distinction and thesis award. The main vision behind Martin’s research is to enable adaptive, robust and sustainable intelligent systems. These lifelong learners are able to behave robustly in novel situations and efficiently include novel experiences, thus providing a sustainable foundation to AI systems for and with humans.
Zoom Registration
Please register with office@c3s.uni-frankfurt.de to receive the Zoom login.