Oberseminar Steven Künzel: Evolving Artificial Neural Networks For Multi-Objective Tasks

26. 06. 2019 | 11.00 Uhr - 12.00 Uhr

hiermit möchte ich Sie herzlich zum Oberseminar

,,Evolving Artificial Neural Networks For Multi-Objective Tasks''

von Herrn Steven Künzel am

Mittwoch, dem 26.6.2019,

um 11:00 Uhr,

im Raum 33/2111 einladen.



Neuroevolution is a growing research field in artificial intelligence and a subdiscipline of reinforcement learning. It aims at evolving artificial neural networks using evolutionary methods. Thereby it combines the research fields of artificial neural networks and evolutionary algorithms. Artificial intelligence represents a huge research field today with applications ranging from cybersecurity to digital games. Among other techniques, artificial neural networks play an important role.

The main contribution of our work is the development of novel neuroevolutionary algorithms based on NEAT -- today's quasi-standard for neuroevolution. Additionally, different techniques for improving the novel algorithms' usability are introduced. Furthermore, we propose the QD measure -- a quality measure that promotes fitness improvement as well as diversity maintenance during the evolutionary process. To that end, we designed a novel algorithm for computing the exact R2 contribution of each solution of a test set which is faster than the original method by factor of the set size. To test the approaches a multi-objective variant of the well known Double Pole Balancing problem as well as a lightweight and fast version of the Fighting Game AI Competition simulator have been developed.

(Der Vortrag findet auf deutsch statt).

Alle Interessenten sind herzlich willkommen!

Termin übernehmen: