German Shepherd "The AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else." - Eliezer Yudkowsky
Query Hinton about non-linear complex manifolds, he will proclaim that you are a CRACKPOT ! But he likes to brag that he only took arts courses as an undergrad, and never a course in computer science, ... or anything to do with physics or advanced math. However, quite good at plying the verbal diarrhea using his British accent, assuming a stance of complete and total authority ! According to Hinton , I am... "DEEP" whist all other minions are "SHALLOW" ! His "DEEP" is little more than a rebranding of the 60 year old multi-layered NN theme. I suppose for him, "SHALLOW" applies to historical figures of Mathematics/Physics... such as Maxwell, LaPlace, Hilbert, Gauss, Newton, Leibniz, Fourier, Lagrange, Euler, VonNeumann, etc, etc, etc...
"having observed truck hitting house and earthquake these are not independent events any more". What about other events that can shake house? Wind, traffic, explosions (gas, ..). These other events not observed (counted) but does not mean they have not or do not occur. So training data must be sufficiently representative of the problem domain. Interesting. Learning to learn. Typing a comment activates some kind of learning mechanism to help think and learn (about learning).
Rupen Patel I don't think current machine learning models have unlocked the metalearning aspect of learning, because the implication of that's just way too scary: sentience in some limited form or another.
I think humans have yet to definitively unlocked the meta aspect of probability. Maybe not the best way to say trying to model chaos with mathematics are doing mediocre job.
50 years from now we look back on this video and say that this guy...this guy started the war of machines.
***** Machines don't have to be evil. I don't understand your question.
German Shepherd "The AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else." - Eliezer Yudkowsky
All these amazing things we can teach machines to do yet we can't teach camera guys to point at the damn slides.
That happens at USC. 😎
How can we bias the network in a particular direction rather than directly introducing network structure?
Is there a part 2?
Masterpiece!!
thanks alot sir... i want real matlab code of any DBN application but not handwriting i need it if you could help me please thanks in advance
The subscriber amount is 666 wtf but this was an amazing lecture.
can the ppt be shared?
www.ipam.ucla.edu/programs/summer-schools/graduate-summer-school-deep-learning-feature-learning/?tab=schedule
Thanks for sharing
thanks for sharing this video :D
p i x - *ELS*
good job
Geoff is incredibly underrated. I hav tonnes of idea to seek his advice on. Please PM me if you're seeing this Geoff!
Query Hinton about non-linear complex manifolds, he will proclaim that you are a CRACKPOT !
But he likes to brag that he only took arts courses as an undergrad, and never a course in computer science, ... or anything to do with physics or advanced math.
However, quite good at plying the verbal diarrhea using his British accent, assuming a stance of complete and total authority !
According to Hinton , I am... "DEEP" whist all other minions are "SHALLOW" ! His "DEEP" is little more than a rebranding of the 60 year old multi-layered NN theme.
I suppose for him, "SHALLOW" applies to historical figures of Mathematics/Physics... such as Maxwell, LaPlace, Hilbert, Gauss, Newton, Leibniz, Fourier, Lagrange, Euler, VonNeumann, etc, etc, etc...
"having observed truck hitting house and earthquake these are not independent events any more". What about other events that can shake house? Wind, traffic, explosions (gas, ..). These other events not observed (counted) but does not mean they have not or do not occur. So training data must be sufficiently representative of the problem domain. Interesting. Learning to learn. Typing a comment activates some kind of learning mechanism to help think and learn (about learning).
Rupen Patel I don't think current machine learning models have unlocked the metalearning aspect of learning, because the implication of that's just way too scary: sentience in some limited form or another.
I think humans have yet to definitively unlocked the meta aspect of probability. Maybe not the best way to say trying to model chaos with mathematics are doing mediocre job.
Very good lecture. I was just trying to understand and learn with my original comment.