Deep Learning Research and the Future of AI [video](microsoft.com)
microsoft.com
Deep Learning Research and the Future of AI [video]
https://www.microsoft.com/en-us/research/video/research-focus-deep-learning-research-future-ai/
9 コメント
It's a pretty great general book that I go back to every now and then when I'm reviewing something I'm learning (i.e. GANs) but isn't something I would recommend for newbies to just pick up and try to understanding without at least spending some time learning math/stats since it's a bit more technical than what non-technical might be prepared for.
There's better introductory material out there to Deep Learning, i.e. fast.ai, Ng's Coursera course, Thrun's Udacity's introductory course, tutorials on Medium, where they explain the math/stats behind something but you can get away with learning the process first and play around with code.
Furthermore for some newbies, I think it's a little easier to understand the material when they try to play with it in notebooks (as is the case of books like Hands-On Machine Learning with Sci-Kit Learn and Tensorflow) than trying to just memorize statements on a page.
But certainly for those here who are more technically orientated, it's an excellent book to pore over. I just like to caution friends interested in Deep Learning that it's not the be all, end all way of getting into it and that if someone out there is interested there are more gradual learning curves elsewhere to get their feet wet before committing to trying to go deeper.
There's better introductory material out there to Deep Learning, i.e. fast.ai, Ng's Coursera course, Thrun's Udacity's introductory course, tutorials on Medium, where they explain the math/stats behind something but you can get away with learning the process first and play around with code.
Furthermore for some newbies, I think it's a little easier to understand the material when they try to play with it in notebooks (as is the case of books like Hands-On Machine Learning with Sci-Kit Learn and Tensorflow) than trying to just memorize statements on a page.
But certainly for those here who are more technically orientated, it's an excellent book to pore over. I just like to caution friends interested in Deep Learning that it's not the be all, end all way of getting into it and that if someone out there is interested there are more gradual learning curves elsewhere to get their feet wet before committing to trying to go deeper.
He says general AI is 'certainly an end goal' but I certainly hope the 'end' part of that is not true. We should certainly expect that the first general AI will share the tremendously significant shortcomings and flaws that human brains give rise to (conflating correlation with causation, rejecting truth on pragmatic bases without determining actual truth or falsehood, etc... every list of 'common logical fallacies' is a flaw that the associative nature of our brain makes feel true despite being (at best) non-determinitive). If we intend to stop there we really might as well not work towards it. I'd expect the end goal to be producing something better.
Same video on YouTube:
https://www.youtube.com/watch?v=5BrNt38OraE
(YouTube seems to stream the video better)
(YouTube seems to stream the video better)
Can anyone bottomline it for us, is this any good? Is there a other state of the union at this moment of late 2017 that you would prefer?
It's like, fine. If you're an engineer curious about the basic notions of deep learning, have at it. It's not very technical; very big picture.
If you're well experienced with deep learning already, maybe only the last 2-3 minutes on the future is worth watching.
If you're well experienced with deep learning already, maybe only the last 2-3 minutes on the future is worth watching.
The interview itself was good but rather high level. My personal bottom line: unsupervised learning is the next frontier.
The future IMO is recursive meta learning. This far the max recursion depth afaik has been 1, and that's just a first step.
Also, SeLUs were supposed to be game changing. Are they, or why are they not? Why haven't they been used more? Or what's still missing?
Also, SeLUs were supposed to be game changing. Are they, or why are they not? Why haven't they been used more? Or what's still missing?
Scaled-ELU is meant to create self-normalizing nets, but that's nothing you can't achieve a bit less efficiently with explicit normalization (batch norm, etc), so hardly a game changer.
http://www.deeplearningbook.org/
Seems to have good reviews on Amazon:
https://www.amazon.com/Deep-Learning-Adaptive-Computation-Ma...