Evolutionary computation is basically zero-th order optimization; in many cases, it is based on
comparisons only. This makes this paradigm convenient when something non-numerical or at least
non-differentiable, such as human preferences, is in the loop. Computer vision is a typically AI
field in which non-differentiability rules and humans are the right evaluation process. It makes
therefore sense to apply evolutionary computation to image generation.
Olivier Teytaud is research scientist at Facebook AI Research in Paris. He has been working in
power systems, arithmetics, games, control, optimization, computer vision. He currently contributes
to Nevergrad (
), a platform for derivative-free optimization, and Polygames (to be open sourced soon), which
recently won the first game against a top level player in the game of Hex. He currently combines
evolutionary optimization and deep learning, including applications in computer vision.
University of Konstanz, Powerwall Room, C202
The talk will be transmitted to Stuttgart, VISUS, Powerwall Room