Hi all,

Tomorrow we'll continue with our zoom group seminar with a talk by Janka Möller, who did an externally supervised thesis on "Deep Reinforcement Learning Applied to Byzantine Agreement". See below for the abstract.

The link to the meeting is https://ethz.zoom.us/j/362994444

Best,

-joe

Abstract: We use Deep Reinforcement Learning to simulate worst-case behaviour of failing nodes for reaching Byzantine Agreement. In particular we study the King algorithm for the synchronous communication model and the Ben-Or algorithm for the asynchronous communication model. We find that the byzantine agent successfully learns strategies leading to maximal run time in various settings of both algorithms and we study the resulting actions in detail. We conclude that Deep Reinforcement Learning is a useful tool to simulate worst-case behaviour in the Byzantine Agreement Problem.