The amount of foresight in that room is incredible. At 38:50 they are basically proposing the neural scaling laws
@diodin85873 ай бұрын
How is NTM different with Transformers? They are both dependent on attention mechanism.
@johnmcdonald45148 ай бұрын
Love this. Thank you
@ultrasound145910 ай бұрын
39:10 not anymore 😂GPT 😊
@VinBhaskara_ Жыл бұрын
cannot stop laughing looking at the smirks on Alex Graves' and Ilya's faces as Schmidhuber talks about his diploma thesis! "Here he goes again!"
@TheAIEpiphany3 жыл бұрын
Great talk, a bit rushed but great nonetheless. I wonder whether we could combine DNC-like machines with program synthesis, the type 2 kind of approach that Francois Chollet is evangelizing. NTM/DNC always have these types of glitches inside where they are almost correct but not quite there.
@beizhou40253 жыл бұрын
Is there any follow-up work on NTM/DNC recently?
@DavenH4 жыл бұрын
Those are extremely cool results. I had the idea to work on something similar to that program "fragment" store for program search, but hadn't followed through. Thank you Juergen!
@WilliamFedus14 жыл бұрын
Mike is hilarious. Great talk!
@vasuwatkittiwangchai60984 жыл бұрын
Thanks for this
@finanstudent4 жыл бұрын
I would love to talk with Alex. I have one, too. I developed it over half a decade ago. See: kzbin.info/www/bejne/aIjQg3l8ds-md9E
@paulwerbos72535 жыл бұрын
test
@ChampikaNirosh5 жыл бұрын
I am from Sri Lanka. I am surprised to see the view count of this video!! don't we get a few million views in a couple of days for a lady gaga's music video? Paul Werbos will be admired one day!!
@ProfessionalTycoons5 жыл бұрын
great video!
@ButterFly-xt2yh6 жыл бұрын
The audio quality isn't good. Anyone knows the paper that Ilya was referring to at 1:01:57
@jonas.eschmannАй бұрын
Should be this one: "Opening the black box: low-dimensional dynamics in high-dimensional recurrent neural networks"
@IgorAherne6 жыл бұрын
Hey guys, I made a visualisation of a Link Matrix and Precedence vector, Check it out: /watch?v=SLNGwmM6JnE
@kunzhan10666 жыл бұрын
:)
@somekid3386 жыл бұрын
Awesome!
@apoozeo17156 жыл бұрын
Why does this talk have only 800 odd views? This is next level tech.
@somekid3386 жыл бұрын
I think this is because the most influential AI groups (DeepMind and OpenAI for instance) put more of their time into non-evolutionary methods. Also most people consult Stanley and Miikkulainen's paper on neuroevolution, though I think this video demonstrates how powerful neuroevolution can be.
@udaywali7 жыл бұрын
Thank you for uploading these lectures. Mind blowing.
@udaywali7 жыл бұрын
Very interesting history of learning algorithms. Future is so strange, where will all this lead us to!
@jammystraub4887 жыл бұрын
Thank you for posting these talks.
@flamingxombie7 жыл бұрын
Thanks so much for posting these videos! I'm reckoning that this is the highway networks guy.
@pardoharsimanjuntak14837 жыл бұрын
What AI thinks is what its creator thinks. The AI creator intends for all humanity to survive, regardless of its condition.
@davincifyable7 жыл бұрын
Where's the hat ?
@machine-learning10137 жыл бұрын
This is a great concept! *Neural Turing Machines* , how *insertcurseword* cool is that !
@theempire007 жыл бұрын
Thanks!
@anyoldthing7 жыл бұрын
learn some new jokes jurgen.
@arturodeza38167 жыл бұрын
His jokes never go old.
@martagutierrez38587 жыл бұрын
Using the same old jokes IS the joke.
@DonatusSchmid7 жыл бұрын
That is kind of deep learning
@arthurpenndragon6434 Жыл бұрын
the repetition is the point.
@vigneshathreya54117 жыл бұрын
Love him or hate him, you've to admit that Schmidhuber has a vast expanse of work