I agree with @tdietterich and I would add the fact that alphago is learning by reinforcement learning, exactly like our brain:https://www.youtube.com/watch?v=Y85Zn50Eczs …
who wants to define deep learning? @tdietterich perhaps you could go first? (and maybe address issues I raised in Medium article, questions 6 and 7? seems like you skipped those re your terminology point)https://twitter.com/tdietterich/status/952707787087228928 …
-
-
-
I'm pretty sure that if our brain is doing reinforcement learning, it is not doing it the same way as AlphaGo. Do you have neurological evidence?
- 16 more replies
New conversation -
-
-
I'm confused by your question. I defined DL as differentiable programming done by people who call themselves deep learning researchers.
-
You ask for some examples of cool unsupervised DL. I think GANs and related ideas are very important. We can now learn high-dimensional density estimators. I believe this will give us generative models for nearly everything (incl. classification)
- 2 more replies
New conversation -
-
-
Indeed, which only bolsters my point. There are many hundreds of thousands (m/billions?) of training iterations at the sub-task level that are done prior to task execution that inform the decision space.
-
Point being that the task of "don't drive off a cliff" is a collection of a million sub-tasks that are optimized and "learned" through something like RL. The combinatory effect of all of them looks like the singular task, so it's simplistic to think of that as a singular task IMO
End of conversation
New conversation -
-
-
Definitely agree that there is some bootstrapped capability there, however we haven't been able to figure out any method of quantifying the impact of epigenetic learning, so I consider it outside of the model.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.