Recent advances in foundation models are extending AI beyond perception and language understanding toward embodied systems that can perceive, reason, and act in the physical world. This transition is accelerating the development of vision-language-action (VLA) models and robot foundation models that aim to provide robots with more general, adaptive, and deployable intelligence.
\r\nThis talk presents recent progress toward self-working humanoid robots, covering key ideas in embodied AI such as multimodal understanding, long-horizon planning, action generation, and real-world deployment. It also introduces recent research from Seoul National University BI Lab and AI Institute on human-robot interaction, personalization, continual adaptation, and language-conditioned robot learning, together with the Habilis framework developed at Tommoro Robotics for deployable humanoid intelligence.
\r\nFinally, the talk discusses a future architecture for self-working humanoids that integrates perception, memory, reasoning, planning, and action into a unified learning system. The central theme is that while large language models enabled machines to communicate, vision-language-action models will enable machines to work autonomously in the real world.
\r\n
\r\nByoung-Tak Zhang's biography
\r\nPresent:
\r\nProfessor of Computer Science, Seoul National University,
\r\nDirector of the Humanoid AI Robot Center (HARC),
\r\nChairman of the K-Humanoid Alliance (Korean Government)
\r\n2019-2025:
\r\nFounding Director of SNU AI Institute (AIIS)
\r\n2007-2008:
\r\nVisiting Professor, Samsung Advanced Institute of Technology
\r\n2003-2004:
\r\nVisiting Professor, MIT CS&AI Lab (CSAIL)
\r\n1992-1995:
\r\nResearch Associate, German National Research Center for Computer Science (GMD)
\r\n1988-1992:
\r\nPhD in Computer Science, University of Bonn
\r\n1982-1988:
\r\nBS and MS in Computer Science, Seoul National University
\r\n
Thesis Directors: Prof. V. Savona, Dr I. Tavernelli
\r\nPhysics doctoral program
\r\nThesis Nr. 12091
\r\n
\r\nTo take part in the public defense, please contact directly the speaker
The talk is jointly organized by the EPFL AI Center and the EPFL Signal Processing Laboratory (LTS5) as part of the AI Fundamentals seminar series.
\r\n
\r\nHost: Prof. Jean-Philippe Thiran
\r\n
\r\nTitle
\r\nDeep Learning Solution of the Eigenvalue Problem for Differential Operators
\r\n
\r\nAbstract
\r\nSolving the eigenvalue problem for differential operators is a common problem in many scientific fields. We introduce a novel Neural Network (NN) -based solver for the eigenvalue problem of differential self-adjoint operators, where the eigenpairs are learned in an unsupervised end-to-end fashion. We propose several training procedures for solving increasingly challenging tasks towards the general eigenvalue problem. The proposed solver is capable of finding the M smallest eigenpairs for a general linear differential operator. Our proposed method simultaneously solves several eigenpairs and can be easily used on free-form domains. We exemplify it on L-shape and circular cut domains. A significant contribution of this
\r\nwork is an analysis of the numerical error of this method. In particular an upper bound for the (unknown) solution error is given in terms of the (measured) truncation error of the PDE and the network structure.
\r\n
\r\nBio
\r\nNir Sochen, finished his B.Sc studies in Physics 1986 and his M.Sc. studies in theoretical physics 1988 both in the University of Tel-Aviv. Received his PhD in Theoretical physics 1992 from the Universit\\’e de Paris-Sud while conducting his research in the Service de Physique Th\\’eorique at the Centre d’Etude Nucleaire at Saclay, France. He continued with a one year research in the Ecole Normale Superieure in Paris on the Haute Etude Scientifique fellowship, and a three years NSF fellowship in the physics department of the University of California at Berkeley. It is in Berkeley that his interests shifted from quantum field theories and integrable models, related to high-energy physics and string theory, to computer vision and image processing. He spent one year in the physics department at the University of Tel-Aviv and two years in the Faculty of Electrical Engineering in the Technion-Israel Institute of Technology. Currently he is a senior lecturer in the Department of Applied Mathematics, University of Tel-Aviv. He is also a member of the Ollendorf center in the Technion.
\r\nHis main research interests are the applications of differential geometry and statistical physics in image processing and computational vision.
\r\n
Are you finding it difficult to come up with a truly innovative idea for your next grant application? Do you ever feel stuck in an experiment, in familiar ways of thinking, even when you know there must be new angles to explore?
\r\n
\r\nScience progress moves between two complementary modes: Day Science, the rigorous testing of ideas, and Night Science, the creative process where those ideas first emerge. Indeed, what we call the scientific method tells us what to do once we have a hypothesis – but where do these hypotheses come from in the first place? François Jacob famously called this part the realm of ‘Night Science’.
\r\n
\r\nDrawing on the Night Science training framework developed by Profs Itai Yanai and Martin Lercher, this workshop will introduce thinking tools and exercises to help you foster exploratory thinking, interdisciplinarity and a general attitude of openness towards new ideas applied to your own research challenges. This course is provided by a certified Night Science trainer from the EPFL Research Office.
\r\n
\r\nThis interactive workshop is designed specifically for postdoctoral researchers who want to:\r\n