Bayesian Optimization for Second-Life Batteries
Ralf Herbrich
Date: Tuesday, Sep 12, 2023, 9:00 - 9:30
Abstract:
How can AI become more energy-efficient and sustainable? And how can AI be used to increase energy-efficiency in other fields, e.g. battery production? In this talk I will present initial ideas, in particular how techniques from Bayesian optimization can be used to infer the state-of-health of second live electrical vehicle (EV) batteries. This is an emerging market with many of the initial EVs coming to the end of their first life while batteries still retain between 70%-80% of their maximum capacity. Using methods of Bayesian optimization combined with electro-chemical models of batteries allows to infer the conditions under which the lifetime of these batteries can be significantly prolonged.
Bio:
Ralf studied at Technical University of Berlin both for a Diploma degree in Computer Science in 1997 and a PhD degree in Theoretical Statistics in 2000, respectively. He has worked in both basic and applied science roles: after a 3-year PostDoc at Darwin College Cambridge, he worked at Microsoft Research from 2000 – 2011, at Facebook in 2011-2012, before leading the core machine learning group in Amazon from 2012 – 2020. Finally, he was the Senior Vice President of AI and Builder Platform at Zalando from 2020 -2022. Since April 2022, he is head of the Artificial Intelligence and Sustainability research group at the Hasso Plattner Institute and University of Potsdam. His research interests include approximate computing, Bayesian inference and decision making, game theory, information retrieval, natural language processing, computer vision, distributed systems, machine learning theory and knowledge representation and reasoning.