[bsaul]

That was also my feeling, and i have it pretty much every time i hear about "deep learning" : it's all PR and people trying to get as much funding as they can while the trend is hot. I have a suscipicion that it will all collapse like a bubble once again.

The only thing that really impressed me so far was that ibm computer playing jeopardy, but the fact that no other application became public after all this years make me wonder if once again all people built was a manually tuned specialized system.

Note : i'm not in the field so my feelings are just based on the communication around the subject and decades of claims about building an intelligent algorithm with no success.

[kylebrown]

Deep Mind Atari demo - http://www.youtube.com/watch?v=EfGD2qveGdQ

[bsaul]

Do we know what reenforcement method they used ? Did the training on one level of breakout had the algorithm perform well on other levels of the same game without any new training ?

Did those games had any kind of random behavior or does the same things happen all the time at the same time ?

It is a progress, i agree, but all those games are just about issueing sequences of "left right" commands to maximize the time spent playing the game.

Things would be a lot different if they could somehow analyze the structure of the network's "conceptual" layer to identify functions over areas ( like " this is where ball trajectory is identified, and we can see it rest and activate depending on the ball's motion" or something similar). But the slide on his presentation shows a big question mark there, which isn't really reassuring.