Ich kopier mal die Praktikumsbeschreibung aus der Mail von Professor Jan Peters:
und noch ein paar interessante Antworten auf Fragen dazu:Project: Solving TETRIS with Policy Search
Supervisor: Gerhard Neumann
Content: Implement and compare several policy search methods to
create optimally learning Tetris players.
Requirement: Some machine learning background, extremely good programming ability.
> Anyway, the first project sounds very interesting to me. Could you give me an example for a policy search algorithm that could be implemented?
Policy Gradient Methods, EM-like Policy Search, Policy Search with the Cross-Entropy Method, Relative Entropy Policy Search
> And would we have to choose them ourselves or do you provide all the papers?
We can provide papers, and help!
> I would also like to know if everybody on the team needs to have a machine learning background or if it is enough if the knowledge is in the team, i.e. how is the machine learning to implementation ratio?
I think it is good but not essential if more members have a good ML background. But trying an algorithm is always the best way to understand it!