skinner_

12 ms·

91.
by skinner_ 7y ago
https://users.renyi.hu/~daniel/tmp/noitdoesnt.jpg
92.
by skinner_ 7y ago
One advantage is that seen.add(line) can be changed to seen.add(hash(line)) which can be significantly more memory efficient for files with long lines.
93.
by skinner_ 7y ago
Loosely related, one of my favorite aphorisms ever: “One man’s vicious circle is another man’s successive approximation procedure.” -- Cosma Shalizi
94.
by skinner_ 7y ago
Complexity theorists do care about reductions within subexponential time classes, but it's not a great topic for P vs NP focused Complexity Theory 101. And when you do focus on P vs NP, the dichotomy is quite natural because of the ht
95.
by skinner_ 7y ago
I bet there are lots of public employees making 99k.
96.
by skinner_ 7y ago
And that is fine by me. Using my behavioral data to train an underfitting model is very different from actually storing my behavioral data. Sure, the word 'underfitting' does the heavy lifting in my previous sentence. But I don&#x
97.
An external advisory council to help advance the responsible development of AI(blog.google)
2 points by skinner_ 8y ago|0 comments
98.
by skinner_ 8y ago
I'd start with https://github.com/mxgmn/WaveFunctionCollapse
99.
by skinner_ 8y ago
I checked this out because I expected to learn about the brand new RT Core accelerated ray tracing. Turns out this tutorial does not touch that topic. It's awesome either way, but https://devblogs.nvidia.com/nvidia-opti
100.
by skinner_ 8y ago
In both cases, there's a low-dimensional latent space, and there's a pixel space of images. In the case of Eigenfaces, the correspondence between the two is linear. That's not a bad approximation, but "make him smile&quo
101.
by skinner_ 8y ago
The seeds contain oil, protein, and a poison. Before the genetic modification, you took the oil, and threw away the valuable protein together with the poison. (Or gave them to cows.) After the modification, you just eat the whole seed.
102.
by skinner_ 8y ago
Besides the already-mentioned pytorch and tensorflow bindings, http://dlib.net/ is an option if C++ is really a hard requirement. It is opinionated and incomplete, basically being a one-man-show, but it has some weird-but-c
103.
by skinner_ 8y ago
More than just commenting. John Baez's blogpost is just elaborating on a math trick invented by Greg Egan. It's a variation on an idea by Hanspeter Schmid.
104.
by skinner_ 8y ago
> Even Tensorflow and Google have acknowledged this, and are trying to rewrite Tensorflow in one language, namely, Swift! Come on.
105.
by skinner_ 8y ago
> It seems to me that the statement that Wu's experiment holds globally is equivalent to the statement that the universe has an orientation so I now tend to think the universe does not admit one of these orientation-reversing loops.
106.
by skinner_ 8y ago
Here's something even crazier: Are photons in vacuum, with photon-photon scattering, IN PRINCIPLE a model with strong enough computational power? I don't have any specific application in mind, but just for the kicks, how cool woul
107.
by skinner_ 8y ago
The epic Turbo Pascal Runtime Error 200 is similar. Turbo Pascal for DOS runtime sleeps by running a no-op loop some number of times. It figures out this number during startup, and it immediately crashes with division by zero if your comput
108.
by skinner_ 8y ago
"DNNs are nothing more than a huge set of NNs" As in deep neural networks are nothing more than a huge set of shallow neural networks? That's directly contradicted by tons of evidence. See for example the visualizations in h
109.
by skinner_ 9y ago
Awesome! You are amazingly fast! I've tried to add neural time warping to it, but it did not help yet. L1 loss on the parameters did help reducing the parameter count on the fancy elephant task. Here are a couple of failed runs: https
110.
by skinner_ 9y ago
That's one way to introduce proof-of-stake.
111.
by skinner_ 9y ago
Now that's the real difference between a native and a non-native speaker: As a non-native speaker, I'd probably put about the right amount of articles in there, but I can't even imagine why dropping them sounds 'cutesy&#
112.
by skinner_ 9y ago
Also, unsupervised. Also, no rollouts. They got rid of a lot of complexity. At this point it looks like a reasonable challenge to write a superhuman Go AI in 500 lines of unobfuscated python.
113.
by skinner_ 9y ago
Indeed! Actually I'm quite surprised that with all the competition and snobbery going on at Michelin-star level restaurants, there's no restaurant that tunes their tasting menu to the guest's individual perception. Even a gen
114.
by skinner_ 10y ago
I did something like this myself: https://github.com/danielvarga/string-art My algorithm uses a bit fancier math: I reformulate the approximation problem as a sparse least squares problem over positive integers, solve
115.
by skinner_ 11y ago
"Using bootstrap to esitmate confidenc intervals for pcoin Interval etsimate of parameter."