In Heilbronn, a group of about 40 PhD students, postdocs, and professors gathered at the “Bildungscampus” in Heilbronn, one of TUM’s (many) locations outside of Munich. The focus of this summerschool was on optimization and machine learning, and how the two fields can come together. Guillaume Dalle freshly introduced us to automatic differentiation, which was surprisingly insightful, even after the many machine learning courses that I had during my studies. He explained the foundations of backpropagation through layers where exact gradients are usually unavailable. Axel Parmentier shared details about their research on machine learning and optimization pipelines. With this approach, he and his coworkers want to train machine learning models end-to-end for predicting the objective of an optimization problem. Leo Baty rounded of this session with hands-on coding. Wouter van Heeswijk (re-)introduced us into the basics of reinforcement learning. Bissan Ghaddar guided us through her recent research on machine learning for improving heuristics for established optimization algorithms.

Toulouse is a magnificent and hot city in the summer. And a great one to meet fellow PhD students from around the world! I was able to join the young game theory community in southern France, learning more about contract theory, prophet inequalities/online optimization, and signaling games. Paul Dütting introduced contracts, a game-theoretical model of a sequential-move game where one player hires another one for getting a job done. How can contracted people be motivated to assert effort? Jose Correa gave a lecture on prophet inequalities and online learning in general. You might have heard of the “secretary problem” before, but there is much more … Penélopé Hernandez and Tristan Tomala dived deeper into signaling games with us. These games are mostly characterized by the idea of a sender (who has private information) and a receiver, who needs to take an action based on the information that the sender provides. The associated workshop gave me, fellow PhD students, and researches from the community the opportunity to present their latest findings. I was very happy sharing my work on Bandit algorithms in games and disucssingmy findings with experienced and curious like-minded. ...