Deep Reinforcement Learning for Learning Machines

  Diesen Termin in den persönlichen Kalender (z.B. Outlook, Thunderbird, Lotus Notes) übernehmen
Startdatum:23. Mai 2017
Startzeit:13:30 Uhr
Stoppzeit:14:30 Uhr
Veranstalter:Fachbereich Informatik
Referent:Martin Riedmiller; Google Deepmind
Ort:S2 02|C205
Beschreibung:

Recently, artificial intelligence has experienced a big boost of interest in both academia and industry and even the general public. A lot of this is due to the success of deep learning methods, i.e., neural networks with a considerable number of hidden layers, that show amazing results in a variety of domains such as computer vision or speech recognition.

In this talk, I will in particular focus on data-efficient reinforcement learning methods. Applying neural networks in the context of self-learning control of dynamic systems has been our research topic for more than 20 years. I will highlight progress in this area by going through several examples in both simulated and real domains, from Atari playing agents to more recent examples in robotics.

 

Biography:

 

Martin Riedmiller has been the leading researcher at using neural networks during both the so-called 2nd neural winter as well as the recent renaissance with the deep neural networks. After a PhD and Post-Doc at the University of Karlsruhe and a Post-Doc at Carnegie Mellon University, he has been professor at Dortmund, Osnabrück and Freiburg, as well as a visiting professor at Stanford University and University of Southern California. Recently, he surprised all of us by moving to Google Deepmind! At Deepmind, he has created a team of researchers that has delivered some unprecedented und previously un-imaginable results of reinforcement learning, ranging from learning Atari games to world class go playing.




Kontakt

Öffentlichkeitsarbeit

S2|02
Hochschulstraße 10
64289 Darmstadt

+49 6151 16-25501
kommunikation(a-t)informatik.tu-darmstadt.de

Hinweise auf weitere lokale Veranstaltungen des Fachbereichs können an events@informatik... gesendet werden.

A A A | Drucken Drucken | Impressum Impressum | Sitemap Sitemap | Suche Suche | Kontakt Kontakt | Webseitenanalyse: Mehr Informationen
zum Seitenanfangzum Seitenanfang