Vabljeni na seminar v okviru projekta SMASH z naslovom »Neuroevolution: Optimizing neural networks with evolutionary algorithms«, ki ga bo v četrtek, 26. 6. 2025, ob 14.00 v predavalnici 4 na FRI.  Predaval bo prof. dr. Kai Olav Ellefsen z Univerze v Oslu.

Predavanje bo v angleškem jeziku.

Opis predavanja: Training Deep Neural Networks with Backpropagation has led to impressive breakthroughs across many different domains in the last decade. Backpropagation calculates gradients of weight updates with respect to a loss term, leading to efficient optimization towards a single objective. However, a parallel field of research, NeuroEvolution, investigates the training of neural networks with Evolutionary Algorithms. Rather than following gradients to maximize an objective function, these algorithms leverage the power of population-based search, with benefits including a diversified search across solution candidates, the possibility to overcome deceptive objective functions, and the ability to search through alternative neural architectures and hyperparameters. In this lecture, Kai Olav Ellefsen will cover the basics of NeuroEvolution: What is it, Why is it useful, and How is it done in practice?

O predavatelju: Kai Olav Ellefsen received the M.Sc. and Ph.D. degrees in computer science from the Norwegian University of Science and Technology, Trondheim, Norway, in 2010 and 2014, respectively. During his Ph.D. studies, he was also a Visiting Researcher with the University of Wyoming, Laramie, WY, USA, for six months, in 2013. He worked as a Postdoctoral Researcher with the Brazilian Institute of Robotics, Salvador, Brazil, from 2014 to 2016; and a Postdoctoral Researcher with the University of Oslo, Oslo, Norway, from 2016 to 2019. Since 2019, he has been an Associate Professor with the Department of Informatics, University of Oslo. His research interests include many topics in the intersection between Artificial Intelligence and robotics, including evolutionary robotics, reinforcement learning, and brain-inspired learning mechanisms.