I am working on a research on adaptive systems in art practice and would like to know of any media art installations or other forms of artworks (apart from music) that make use of reinforcement learning. The only examples I have been able to find thus far are works that I have authored or co-authored.
Sofian
I guess it depends how strictly you define 'reinforcement learning'.
In the music realm, was George Lewis's Voyager?
Does David Rokeby's Giver of Names count?
I'd ask Alan Dorin, Joan Soler and Nell Tenhaaf, among others. Also Noah Wardrip Fruin and particulaly Mike Mateas should be helpful. Jens Hauser too.
SP
Hi Simon,
RL has been used for automated / improvisational composition for a while. Music is a good candidate for that because it's a time-based system with a limited number of symbols and data is easy to generate or acquire. Lewis' work does not seem to use any machine learning techniques (from reading his 2000 paper on his work).
Rokeby's Giver of Names (and some of his other works) use pattern recognition and probably (in that case) Markov Chains. I'll dig deeper but I doubt he uses RL. Thanks for pointing me to him though, cause there is surely quite a few pieces of his that use Neural nets or other forms of pattern recognition techniques from Machine learning.
Thanks for the references, I'll try to contact these researchers.
Sofian
We used a very simple RL algo in order to add some adpatibility capability to the iCub playing DJ with a human partner which could give feeback about the music quality.
(see the video attached). We never really investigated the effect of this adaptation though...
http://youtu.be/6wK0Ld13US8
Merci Stephane, quite an interesting project. What was the RL algo used exactly? Did you publish a paper that describes the approach?
- Badges
- Science topic