curuinor

56 ms·

481.
by curuinor 11y ago
This may not be the answer you were hoping for, but if you do not have the background already to do this sort of thing and if you are not already stubborn enough to start learning mathematics on your own, school may be for you. I assume you
482.
by curuinor 11y ago
And if you want to solve SAT with neural networks, you usually go back to a modified version of the Hopfield network. Rumelhart was also inspired by poking at MAX-CSP, at least according to McClelland.
483.
by curuinor 11y ago
Hopfield was the first to poke at Ising model spin to create memory-having systems, creating his Hopfield model where memory states are represented as attractors in the Hamiltonian search space. Add hidden units to that, and you get the gen
484.
by curuinor 11y ago
It's notable that there are actually no enforced prerequisites for 229, besides basically being OK at math. Which meant that I knew an education major who sat down, tried to take it and sort of crashed and burned in the third week.
485.
by curuinor 11y ago
I took the in-person course at stanford. I recall the first problem set in cs229 was about as hard as the entirety of the Coursera course combined...
486.
by curuinor 11y ago
I used to save my ideas, quite rigorously. Now, I try not to. Ideas are nothing, execution is everything, and execution is very much composed of things that you will remember, because they are so specific.
487.
by curuinor 11y ago
No, because the distributedness of the representation means that exponentially large Markov chains would be needed to replace linearly growing deep models. See this paper on the recurrent temporal RBM ( http://www.cs.utoronto.ca&#
488.
by curuinor 11y ago
This is the Chaitin of Kolmogorov-Solomonoff-Chaitin complexity, if the argument seems familiar. I like Grassberger-Crutchfield-Young ( http://www.scholarpedia.org/article/Complexity , look for Statistical complexity) co
489.
Metafilter reinstates Gopher server(metatalk.metafilter.com)
10 points by curuinor 11y ago|0 comments
490.
Show HN: Stupid MLP Sparsification(github.com)
1 points by curuinor 11y ago|0 comments
491.
by curuinor 11y ago
Mandelbrot and NN Taleb's claim requires non-Markovianness of the walks. Fractional brownian motion, for example, exhibits long-range dependence. Also "randomness" is measured using Kolmogorov complexity, a quantity which is
492.
by curuinor 11y ago
Positive feedback loop, specifically. Tesauro noted power-law times in BP learning and you can go off and look at highly skew distributions in BP weight assignments on your own. Simon had a little program of research on skew distributions i
493.
by curuinor 11y ago
H. Simon was among the founders of artificial intelligence and a very important microeconomicist, and I have often thought that these disciplines are not fundamentally different. What problem does backpropagation solve? Credit assignment. W
494.
by curuinor 11y ago
I wrote a thing previously noting that some of the same fractal phenomena happen in word2vec ( http://howonlee.github.io/2016/02/05/Fractal-20Wordvecs.html ).
495.
by curuinor 11y ago
A challenge: find a symbolic system which does not obey a Zipfian distribution or has a heavy tail. You will fail. Programming included.
496.
Fractal Structure in Word2Vec Embeddings(howonlee.github.io)
3 points by curuinor 11y ago|0 comments
497.
Social Network Patterns in Neural Network (Multilayer Perceptron) Gradients(howonlee.github.io)
3 points by curuinor 11y ago|0 comments
498.
Poking at causation, 1 of 3(howonlee.github.io)
3 points by curuinor 11y ago|0 comments
499.
by curuinor 11y ago
Read many children's books. It may be of use, just because of the farness of the language from English and of the effort you will need to expend to learn anything, to get formal schooling, and to use formal textbooks. Yonsei University
500.
by curuinor 11y ago
it's the CDG's library, there's other people in there and it's definitely not just 2015
501.
by curuinor 11y ago
The actual space is really really cool and is the instantiation of their longer-term project, which is supposed to be an attempt to reify D. Engelbart's vision for what computation is to be.
502.
by curuinor 11y ago
No, I'm afraid not. He didn't seem incredibly eccentric to me, anyhow, just a ridiculously hard worker
503.
by curuinor 11y ago
Primary investigator-like people. Like, he hires researchers to work under him.
504.
by curuinor 11y ago
He's one of the PI-like people at the CDG. ( http://www.fastcodesign.com/3046437/5-steps-to-recreate-xero... )
505.
by curuinor 11y ago
Are any of the CDG kids sharing the computer-that-fits-in-a-room dealie they have in this library?
506.
by curuinor 11y ago
The quality of being realistic in a mediated environment is called telepresence, or presence. There's a journal for it. It's a small journal, although pretty old for a CS journal (not at all old for a social science journal) so I
507.
by curuinor 11y ago
I was in Jeremy's lab as an undergrad and master's student. There was a team of student RA's which was from the communications department and a team of student programmers who were CS, EE and Symbolic Systems. Many of the gra
508.
by curuinor 11y ago
And I have also thought about the process of making friends for many years. Consider that the complex network which is considered the ego network (a la the sociologists) is really considered a complex network, just like the overall global s
509.
by curuinor 11y ago
I have implemented and used a similar thing for many years. I think a very great virtue of such a system is that you can run a cronjob pretty easily to email you reminders to contact any friends you haven't poked at much.
510.
by curuinor 11y ago
A. Teller is probably more interesting to HN folks because he's head of Google X. This seems incongruous a little bit and then you see how people in Palo Alto parent...

More ›