one-more-minute

9 ms·

31.
Dancing Plague of 1518(en.wikipedia.org)
3 points by one-more-minute 7y ago|0 comments
32.
New Robot Makes Soldiers Obsolete(youtube.com)
2 points by one-more-minute 7y ago|1 comments
33.
by one-more-minute 7y ago
Anyone not familiar with James Mickens should read "The Night Watch" immediately. https://www.usenix.org/system/files/1311_05-08_mickens.pdf
34.
Ching Shih(en.wikipedia.org)
3 points by one-more-minute 7y ago|0 comments
35.
by one-more-minute 7y ago
This is less a problem with differentiable rendering than a problem of defining image distance, which I agree is hard (though there are plenty of interesting approaches, like losses defined by conv nets). More interesting though is when the
36.
by one-more-minute 7y ago
In a more traditional SCT AD like Tapenade, there isn't really anything that corresponds to recording of program operations at runtime. Zygote arguably does something semantically equivalent to recording a trace (at the interprocedural
37.
by one-more-minute 7y ago
The really big problem with tracing, `@jit` annotations, and similar approaches, is that you're no longer running Python, but something almost-but-not-quite Python – modifying the language's semantics is necessary to get performan
38.
by one-more-minute 7y ago
> As an aside, a pet peeve of mine is people calling these algorithms "tape free". The common autograd-style tape combines a recording of both the operations of a program and its intermediate values, making the term ambiguous;
39.
Continuations in Julia(github.com)
2 points by one-more-minute 7y ago|0 comments
40.
by one-more-minute 7y ago
Python uses arbitrary-precision integers by default, unless you use Numpy, in which case you end up with the same issue. It also uses machine floats by default, which can give similar errors (so one could say "Python doesn't do re
41.
by one-more-minute 7y ago
It's true that most autodiff systems would behave this way, but a human given only the rules of calculus (the same rules as AD) would have trouble at the same point. Of course, most mathemeticians would apply additional insights while
42.
Differentiable Programming(en.wikipedia.org)
2 points by one-more-minute 7y ago|0 comments
43.
A Julia Interpreter and Debugger(julialang.org)
351 points by one-more-minute 8y ago|96 comments
44.
TensorFlow Eager(arxiv.org)
4 points by one-more-minute 8y ago|0 comments
45.
Beating Reinforcement Learning with Julia(fluxml.ai)
3 points by one-more-minute 8y ago|0 comments
46.
Reinforcement Learning vs. Differentiable Programming(fluxml.ai)
10 points by one-more-minute 8y ago|0 comments
47.
Meta-Learning with Julia and Flux(domluna.me)
6 points by one-more-minute 8y ago|0 comments
48.
by one-more-minute 8y ago
Not at all. Only a small number of the people involved in this actually have PhDs (often in unrelated fields) and many of the most prolific contributors are undergrads who get involved via summer projects. If you're interested in inter
49.
by one-more-minute 8y ago
That's essentially what a source-to-source AD does, just with support for the extra features that show up in programming languages. For example, handling variable bindings gets you the typical Wengert list, and handling function calls
50.
by one-more-minute 8y ago
Thanks for the kind words! I'm lucky to be part of such a smart and driven community – there's a really strong shared sense of how ML should look in Julia. I'm always happy to hear about how this stuff is being used, if peopl
51.
What Is Differentiable Programming?(fluxml.ai)
133 points by one-more-minute 8y ago|41 comments
52.
by one-more-minute 8y ago
Historically ML frameworks have only supported simple combinations of matrix multiply and convolution; it's really about generalising the idea of a model to anything that can mathematically be differentiated [1]. For example, an ODE so
53.
by one-more-minute 8y ago
I think Asimov knew all this – his stories are about how the laws can be twisted into having surprising consequences. It's meant as a Sci-Fi version of the "literal genie" [0]. Has anyone actually taken them as a serious sugg
54.
Differentiation for Hackers(github.com)
1 points by one-more-minute 8y ago|0 comments
55.
by one-more-minute 8y ago
It's specific to CPS form: every basic statement is a full function call, which moves to the next statement by calling another function (i.e. the continuation). Normally this would just be a very quick way to get yourself a stack overf
56.
JAX: Numpy with Gradients, GPUs and TPUs(github.com)
132 points by one-more-minute 8y ago|29 comments
57.
Bézier moi(cormullion.github.io)
190 points by one-more-minute 8y ago|35 comments
58.
by one-more-minute 8y ago
Yes, and unlike AD and GPU support, things like logging have nothing (special) to do with ML. Julia has both very nice logging and plenty of good serialisation options, all of which works nicely with the ML stack. It's entirely unneces
59.
Building a Language and Compiler for Machine Learning(julialang.org)
161 points by one-more-minute 8y ago|24 comments
60.
SIMD in Julia(kristofferc.github.io)
20 points by one-more-minute 8y ago|0 comments

More ›