orange3xchicken

9 ms·

31.
by orange3xchicken 6y ago
imo, instead of developing new theory entirely from the ground up, its more useful to address & work in the context of understanding longstanding open problems/phenomenon. Theoretical insight and the framework should follow. e.g. R
32.
by orange3xchicken 6y ago
I'm pretty impressed at how frequently Yuanzhi Li & Zheyuan Allen Zhu are able to publish high-quality dl theory papers.
33.
by orange3xchicken 6y ago
A lot of valid criticism of this post on the reddit ml subreddit: https://www.reddit.com/r/MachineLearning/comments/kqazpd/d_w...
34.
by orange3xchicken 6y ago
Yeah - I looked it up. Thanks!
35.
by orange3xchicken 6y ago
Yikes! This doesn't really come across as a nice comment (bent minds???). There are good concrete reasons to pursue a PhD (ignoring soft reasons like pure interest): wanting a research career is one - it's pretty difficult to get
36.
by orange3xchicken 6y ago
They do rely on english & spanish translations to construct the dataset. They have existing tools to jointly embed sentences from multiple languages (laser). These models are trained using parallel corpora involving either english or sp
37.
by orange3xchicken 6y ago
I mean it really depends how deep you want to go. Like you point out that the classes you took are 101 courses. These are really just "tasting" courses. I'm sure if you decided to take more advanced/grad numbered courses
38.
by orange3xchicken 6y ago
If you are at all concerned with performance and can sacrifice a couple features (fewer supported text formatting options & mathematical elements, less legacy support for some browsers). Katex is a great alternative. https://
39.
by orange3xchicken 6y ago
I'm not sure I agree. They have taken some liberties with the dialogue, but many of the quotes in the trailer are near-direct from the novel. Anyways this seems like something that would be hard to tell from what little the trailer off
40.
by orange3xchicken 6y ago
In contrast, I have really enjoyed & have found many things likable abt all his movies. Also worth noting that afaik, all his movies post-Arrival just consists of BR2049.
41.
by orange3xchicken 6y ago
Conversely, in theoretical cs, there are many longstanding open problems. People who make progress on these problems are taken seriously and listened to precisely because they have new techniques and novel approaches that have not been cons
42.
by orange3xchicken 6y ago
Many mathematicians would disagree with the claim that automated proof checking/generation (see COQ & this presentation by Voevodsky: https://www.ias.edu/ideas/2014/voevodsky-origins ) and involved techniq
43.
by orange3xchicken 6y ago
I agree with this, and this was true historically as well - e.g. Biometrika - a premier journal in statistics - grew out of bioinformatics-like research. Much of the groundwork laid for online learning & statistics/bandit algorithm
44.
by orange3xchicken 6y ago
This is a good comment, but I suggest you make some paragraphs! It's a bit hard to parse atm. To add on to what you said, many academics pursue knowledge for its own sake. In many cases, especially in theory, the value of new knowledge
45.
Ron Graham passed away earlier this evening at the age of 84(reddit.com)
4 points by orange3xchicken 6y ago|2 comments
46.
by orange3xchicken 6y ago
Prof Graham's juggling: http://www.math.ucsd.edu/~ronspubs/pro_07_peripatetic.pdf
47.
by orange3xchicken 6y ago
The best thing I remember from Demmel's book is that i'm pretty sure the spooky possessed baby on the cover is his kid.
48.
by orange3xchicken 6y ago
The two classic graduate texts for numerical linear algebra are Matrix Computations by Golub and Van Loan and Numerical Linear Algebra by Trefethen and Bau. The reference you are looking for (includes executable C++) is Numerical Recipes by
49.
by orange3xchicken 6y ago
Chrome has recently (within the last year) been repeatedly making me frustrated with its pdf issues. It feels like 40% of the time I open a pdf in chrome, I instead get an error message like "the plugin has crashed" or some bs. Th
50.
by orange3xchicken 6y ago
I just finished The Nothing That Is - it was alright. The illustrations were cool, but the writing and content were only okay imo. I really enjoyed Fermat's Last Theorem. I've also heard good things about - Zero: The Biography of
51.
by orange3xchicken 6y ago
This is a pretty good paper, & they bring up many reasonable points, but I think it's important to distinguish deep metric learning from more traditionally formal ml methods for metric learning - there is plenty of progress being m
52.
by orange3xchicken 6y ago
Thanks - that's great! I'm familiar with Kolter's work on adversarial robustness.
53.
by orange3xchicken 6y ago
In general, I think neural networks really struggle with np hard problems. Current state of the art is bottlenecked wrt to even subtle generalization. The strong baselines have been variations on the pointer-network architecture for some ti
54.
by orange3xchicken 6y ago
It looks like they go from tree -> sequence via prefix notation. I'm curious why Lample decided on this seq2seq approach when it seems that there might be models which could be more naturally applied to the tree structure directly [
55.
by orange3xchicken 6y ago
Yeah, the original work this paper follows up (by the same group: https://arxiv.org/abs/1803.03635 ) received a lot of attention in 2018 when it was uploaded to Arxiv & spawned a lot of followup work. Even though it
56.
by orange3xchicken 7y ago
I think it's relatively common - I guess it also depends on how sensitive the cause of death is. There is a pretty good nytimes doc on obit journalists: https://www.youtube.com/watch?v=G2YXpyh_YOw
57.
by orange3xchicken 7y ago
The theorem you mention is true for networks whose widths tend towards infinity. This paper assumes a nn is given with fixed width n and fixed depth l. The main result is that there exists a subnetwork of a nn with depth 2l and width polyno
58.
by orange3xchicken 7y ago
Just re spectral methods on graphs: There are a number of really interesting interpretations of spectral methods including energy stabilization of spring networks and discrete analogues to heat diffusion over grids. The area is really fasci
59.
by orange3xchicken 7y ago
I always thought of wikipedia as an Encyclopedia Galactica - even though it's edited more like the Hitchhikers Guide.
60.
by orange3xchicken 7y ago
Legit shops might employ DL algorithms as a way to recover behavioral indicators, but no one is relying on DL or using DL for decision-making.

More ›