Abstract: In this seminar, I will talk about our recent experiment that uses deep reinforcement learning to control the manipulation of an extremely tiny object: an atom. Atom manipulation is a powerful tool in physics research to create materials that don’t exist in nature and devices much smaller than present technology. However, different microscopic phenomena make atom manipulation a stochastic control problem.  Our experiment shows that through thoughtfully formalizing atom manipulation into a reinforcement learning framework and combining several state-of-the-art algorithms, deep reinforcement learning models can be efficiently trained using only real-world data to manipulate atoms with optimal precision. Our results demonstrate that deep reinforcement learning can offer effective solutions to real-world challenges in nanofabrication and increasingly complex scientific experiments 

Speakers: I-Ju Chen is a postdoc at FCAI and Aalto university’s applied physics department. Her research is focused on the intersection of machine learning and experimental condensed matter physics. Before coming to Finland, she was a postdoc at the University of Copenhagen, and she received her Ph.D. in 2019 from Lund University.

Affiliation: Aalto university

Place of Seminar:  Zoom