chillee

36 ms·

331.
by chillee 9y ago
Generalization performance is typically defined in roughly this manner. Assume your training data consists of a sample S of m elements from your full distribution D. Let L(S) be the average loss for your model (hypothesis is also a common t
332.
by chillee 9y ago
This guy has them right. I'd probably argue that CVPR is slightly more prestigious than ECCV/ICCV and NIPS is slightly more prestigious than ICLR/ICML, but deciding which conference to submit to is mostly related to "whe
333.
by chillee 9y ago
Yes. Specifically, the common argument has become that flat minima generalize well. As for the last part, when people talk about overfitting we're typically ignoring distributional shift. Otherwise, models can achieve very good general
334.
by chillee 9y ago
I don't think people actually use Vickrey auctions in practice (although it's likely to use some of the same ideas). Even though it has some really cool properties (you're theoretically perfectly incentivized to bid your real
335.
by chillee 9y ago
Really don't think that's the best paper to say "sheds quite a bit of light on this". That paper has been somewhat controversial since it came out. I think https://arxiv.org/abs/1609.04836 is semina
336.
by chillee 9y ago
I don't think it's definitely true that will work well. AlphaZero did significantly better than the original versions of AlphaGo (which did learn from existing human games). However, even training those nets will still take a fair
337.
by chillee 9y ago
"a while" is a bit of an understatement. Leela zero (the main alphago zero replication project) is a crowd sourced computation effort that's going to take a fairly long time to get anywhere. And from this paper: > "Tr
338.
by chillee 9y ago
This kind of stuff is pretty tricky. If you only account for average human suffering, not only do you not account for happiness but you fall into the trap of concluding that it's best to kill all people suffering.
339.
by chillee 9y ago
He used to be the PM of youtube video classification too.
340.
by chillee 9y ago
I think you have a misconception of what capsule networks are. They are not intended to address "performance issues in really advanced models", they are intended as another paradigm in deep learning that Geoff Hinton thinks has a
341.
by chillee 9y ago
Just another point of data with compton compositor on Arch with several text editors. https://news.ycombinator.com/item?id=14800713
342.
by chillee 9y ago
I saw this book on HN several years ago, and thought it was pretty interesting, and maybe worth reading. Unluckily, it didn't help me very much in learning more about this field, nor did I think it was very helpful for solving other pr
343.
by chillee 9y ago
> I have a similar use case in mind, so I'm planning on trying to get this working by writing a simple script that proxies language server messages over SSH/TCP. Ideally it should work with any language server. I think I know a
344.
by chillee 9y ago
If anybody's curious about the math, it's pretty simple. Assume p is the probability of flipping heads. Then, p p is the probability of flipping 2 heads, (1-p) (1-p) is the probability of flipping 2 tails, (1-p) p is the probabili
345.
by chillee 9y ago
Didn't realize that. In that case, sounds awesome!
346.
Visual Studio Code at Connect();2017(code.visualstudio.com)
3 points by chillee 9y ago|0 comments
347.
by chillee 9y ago
This is incredibly not promising (in regards to pricing): https://twitter.com/VisualStudio/status/930885113159802886
348.
by chillee 9y ago
Personally, I definitely would. Disk space is not really a particularly limited commodity for me, and considering that VSCode's install footprint is already 200MB+, it's not really a problem. Of course, it's always possible t
349.
by chillee 9y ago
The current work for this is being done here: https://github.com/chillee/vscodeneovim
350.
by chillee 9y ago
If you try opening up dev tools when that happens, it's likely the issue will pop up. Honestly, there's several places that from an engineering quality standpoint bother me quite a bit. Performance on rapidly repeated actions is o
351.
by chillee 9y ago
Is "Continue to invest into API that enables remote development" not what you want?
352.
by chillee 9y ago
The problem with `cmdline` modes is definitely that tab completion doesn't work imo. Unluckily, that's blocked on the VSCode side, although it is possible to move our cmdline down to the status bar (probably would have been the be
353.
by chillee 9y ago
I suspect what you're running into is some kind of crash when running a command. I will say that I don't think we're doing a very good job handling errors, and I suspect it's that most users don't report them (they
354.
by chillee 9y ago
You might have seen this already, but there was a thesis on combining subtyping and type inference here: http://www.cl.cam.ac.uk/~sd601/papers/mlsub-preprint.pdf
355.
by chillee 9y ago
To be clear (as the guy behind these efforts), what was merged was a PR to let Neovim handle Ex-commands. This by itself was already pretty cool. In <200 lines of code, we were able to obtain full Ex-mode support, an aspect that had alre
356.
by chillee 9y ago
Not really relevant to the article itself, but I've always been surprised how much attention Facebook's Haskell usage gets compared to Facebook's Ocaml usage. From interning at Facebook, the only project that I'm aware o
357.
by chillee 9y ago
Researchers definitely do not prefer Scikit. If anything, Scikit is a better option than tensorflow for "mainstream tech workers from a corporate environment".
358.
by chillee 9y ago
Visual Studio Code team and Visual Studio team are completely separate, AFAIK.
359.
by chillee 9y ago
Could you elaborate on what you mean by "the vim keybindings acting up"?
360.
by chillee 9y ago
A big thing that I use is that if you do quick open (`ctrl+p`) and then type `term `, you can then type the terminal number to go to the specified terminal.

More ›