Stack Exchange
log in

Sasho Nikolov

 Discussion on question by SomeOne: Is…

Imported from a comment discussion on academia.stackexchange.c...
Sasho Nikolov
Dec 3, 2018 14:50
Starting your obnoxious rant with a glaring typo ("I'he") in big bold blue letters is also pretty stupid. Maybe I should link to this guy's page from mine, under "most ridiculous faculty pages I have seen".
 

 Discussion on question by Richard Bro…

Imported from a comment discussion on academia.stackexchange.c...
Sasho Nikolov
Sep 13, 2018 20:04
Can you slack your way through a PhD? Absolutely! Can you get a tenure track professor job this way? Very very unlikely. So I think if you want to be a professor you are asking the wrong question. The academic job market is tough and meeting bare minimum requirements of the job is very far from sufficient.
 

 Discussion on question by Emanuele Vi…

Imported from a comment discussion on cstheory.stackexchange.c...
Sasho Nikolov
Sep 3, 2018 05:56
@SinanÜnür even Strassen's algorithm is barely used in practice - it's not easily parallelizable, has poor locality, and is worthless for small matrices. So, in terms of success stories, page rank is way more successful than fast matrix mult., no matter how beautiful the latter is. (Also your logic makes no sense to me - there are many ways to do something, and they're not all created equal.)
Sasho Nikolov
Sep 3, 2018 05:56
I think there is something slightly problematic about the way you ask your question. The combination of "big success story" and "taught in standard algorithms courses" leads to algorithms like Simplex and FFT. Both of them are really widely used, but they've become so standard, that it's hard to find specific references. It's almost like finding a reference for someone using binary search
 

 Discussion on question by Warren Schu…

Imported from a comment discussion on cstheory.stackexchange.c...
Sasho Nikolov
Sep 2, 2018 21:42
While this question has some value in coordinating the discussion going on in other places, I think it also makes it very obvious why SE is not a good discussion format.
 

 Discussion on question by domotorp: C…

Imported from a comment discussion on cstheory.stackexchange.c...
Sasho Nikolov
Dec 26, 2017 20:22
@Saeed I have only skimmed the paper but I do not think their NNs output a vertex cover. The NN outputs an evaluation function, used for choosing the next element to add to a solution in a greedy algorithm. The evaluation function itself is a very simple algorithm, parametrized in some way, which can be run as a part of a greedy algorithm on other instances. It's all quite constrained but that's not surprising.
Sasho Nikolov
Dec 26, 2017 20:22
I just found out this exists: arxiv.org/abs/1704.01665
Sasho Nikolov
Dec 26, 2017 20:22
This seems underspecified: "For example the network would get an input graph G with weighted edges, and two vertices ss and tt specified, and we asked it to find a shortest st path as fast as possible". You don't "ask" a neural network to do something in this way. You train it on some data. What would the data be in this case? Examples of graphs with correctly labeled shortest paths? Generated from what distribution? You can only expect the network to perform well on the same distribution.
 

 Discussion between Sasho Nikolov and …

Imported from a comment discussion on cstheory.stackexchange.c...
Sasho Nikolov
Nov 19, 2017 03:53
This should be made explicit in your pseudocode. On the other hand the fact that X is a TM and X(X) does not halt is a red herring
Sasho Nikolov
Nov 19, 2017 03:53
To do this, you need to know which machines halt, so that you can simulate only the ones that do halt. Otherwise your simulation would never halt.
Sasho Nikolov
Nov 19, 2017 03:51
You need the halting problem to be solvable in order to test if there exists a machine M and input I with |M| + |I| <= N such that M outputs X
Sasho Nikolov
Nov 19, 2017 03:51
The argument uses the solvability of the halting problem in a different place, and you do not make that explicit, which contributes to why your answer is confusing
Sasho Nikolov
Nov 19, 2017 03:50
Threre is also no place in the argument where you need X(X) not to halt
Sasho Nikolov
Nov 19, 2017 03:50
X does not need to be a Turing machine: absolutely nothing in the argument changes if you treat it just as a string.
Sasho Nikolov
Nov 19, 2017 03:47
As much as I enjoy arguing about my own ridiculousness on the Internet, I decided to improve your answer instead. Feel free to roll back my edit if you don't like it. Notice that $X$ did not need to be a Turing machine, and this is not where the Halting problem was needed in order to make $f$ computable. Do you disagree that the fact that this argument can be rephrased to follow from Roger's fixed point theorem suggests that it's not really fundamentally different from the diagonalization proof?
Sasho Nikolov
Nov 19, 2017 03:47
BTW I think your answer is not clearly written. Even though I know this proof, I still had trouble understanding what you wrote. The pseudocode is redundant and can be taken out, while the paragraph starting with "Now we set" is unclear and needs much more detail.
Sasho Nikolov
Nov 19, 2017 03:47
I think there is self-reference, but it's a little buried: a key step in the proof is that if $N$ is big enough, the function $f$ must return its own code, and then this leads to a contradiction. I think the fact that the argument can be captured by a variant of Lawvere's fixed point theorem (see cstheory.stackexchange.com/a/37830/4896) is evidence that it's really the same argument at heart as other self-referential proofs.
 

 Discussion on question by Hakaishin: …

Imported from a comment discussion on academia.stackexchange.c...
Sasho Nikolov
Jun 16, 2017 14:43
I would be very surprised if there is all that much time or money spent on computing S(5).
 

 Discussion on answer by Anonymous Mat…

Imported from a comment discussion on academia.stackexchange.c...
Sasho Nikolov
Jan 1, 2017 01:00
A problem sitting unanswered on MO for 1.5 years with 5 upvotes could also indicate lack of interest and/or frustration with your notation or with your many edits, among many other things. Several comments on your MO question suggest that you need to familiarize yourself with some modern developments in harmonic analysis. I second the suggestion to put this problem on back burner, go to grad school with an open mind, and consider coming back to it once you have picked up more tools.
 

 Discussion on answer by ff524: How to…

Imported from a comment discussion on academia.stackexchange.c...
Sasho Nikolov
Dec 12, 2016 19:29
@robertbristow-johnson in pure math it's quite common to get a single review. i had a paper accepted in the proceedings of the ACM with a single review, and that's a decent enough journal.
Sasho Nikolov
Dec 12, 2016 19:29
@BioGeo so you are basically saying that if an editor does not agree with your review and decides to ignore it, they must be playing politics.
 

 Discussion on question by anon. compu…

Imported from a comment discussion on cstheory.stackexchange.c...
Sasho Nikolov
Sep 28, 2016 00:33
just to re-affirm Ricky Demer's point, a Riemann integrable function could be discontinuous on an infinite set, even an uncountable one. For example, the indicator function of the Cantor set is Riemann integrable.
 

 Discussion on answer by ff524: Would …

Imported from a comment discussion on academia.stackexchange.c...
Sasho Nikolov
Aug 22, 2016 02:09
@iayork A professor who is an expert in the field in which the paper is written can speak about why the paper is important, what is innovative about it, how influential it's likely to be. This information cannot be gleaned from just knowing that a journal is highly ranked: there is variability among papers in prestigious journals as well. That's why departments ask for tenure case letters from experts who have not worked with the candidate.
 

 Discussion on question by Turbo: NP c…

Imported from a comment discussion on cstheory.stackexchange.c...
Sasho Nikolov
Dec 13, 2015 11:08
I think @MohammadAl-Turkistany is thinking of the wrong problem. The problem in the question has a single equation over the reals, rather then multiple equations over GF(2).
Sasho Nikolov
Dec 13, 2015 11:08
@MohammadAl-Turkistany kOnes is defined with more than one linear equation, no?
 

 theory salon

theoretical computer science. highlight reel vzn1.wordpress.co...
Sasho Nikolov
Oct 5, 2014 21:40
@vzn Thesis will be online soon, thanks for the interest.
Sasho Nikolov
Feb 8, 2014 22:01
all good industrial research labs hire almost exclusively PhDs, and getting a job in the really good ones is not any easier than getting a faculty job (although the qualities that you need to show may differ)
Sasho Nikolov
Feb 8, 2014 21:59
then there are places that do fascinating research, but all research must link in a clear way to the company's business, google research is an example
Sasho Nikolov
Feb 8, 2014 21:58
(and in more "pure" research labs it still looks good if your work leads to or can lead to viable products)
Sasho Nikolov
Feb 8, 2014 21:58
there are labs where you can do basic research, and you are also expected to spend a fraction of your time on either consulting or projects that are more directly linked to products
Sasho Nikolov
Feb 8, 2014 21:57
an example would be Microsoft Research
Sasho Nikolov
Feb 8, 2014 21:56
@anorton there are industrial research labs where you can do basic research, of the same kind as in a university department (but without the students, for better and for worse).
 

 Theoretical Computer Science 2014 Ele…

Open Discussion for cstheory.stackexchange.com/election/2
Sasho Nikolov
Jan 27, 2014 17:09
So long
Sasho Nikolov
Jan 27, 2014 17:09
So take it as a one-time feedback.
Sasho Nikolov
Jan 27, 2014 17:08
I meant "I am not saying this to look for a discussion"
Sasho Nikolov
Jan 27, 2014 17:08
are not willing to strikes me as futile.
Sasho Nikolov
Jan 27, 2014 17:08
In any case, I am saying this to look for a discussion, since discussing someone's behavior and trying to argue them into changing it if they
Sasho Nikolov
Jan 27, 2014 17:05
I think your meta question was good and I upvoted it, btw. And while I agree with Dilworth's answer, I do not think it's an answer to the question asked.
Sasho Nikolov
Jan 27, 2014 17:03
Let me emphasize that the only reason I am saying this is in the hope you'd take it constructively.
Sasho Nikolov
Jan 27, 2014 17:02
Even though sometimes there is some usefulness to what you write, the style, which makes your writing difficult to parse, the lack of any rigor, and the general handwaviness make your writing similar in quality to typical crank material, and this imo puts people off.
Sasho Nikolov
Jan 27, 2014 17:00
Your questions have a similar quality of being based on superficial associations rather than formal mathematical statements.
Sasho Nikolov
Jan 27, 2014 17:00
@vzn It's not just your writing style that puts people off (although you do have many writing habits that I find outright annoying). The problem is that your input in the form of questions and answers exhibits the same carelessness as your writing. Many of your answers are either only tangentially related to the question, or the result of an easy Google search, or both, and in any case are not useful.
2
Sasho Nikolov
Jan 24, 2014 19:48
good talking, goodbye
Sasho Nikolov
Jan 24, 2014 19:48
good luck, and if you write a meta question, please try to sound less..alarmist...
Sasho Nikolov
Jan 24, 2014 19:48
ok anyways, i am sorry, i really should go
Sasho Nikolov
Jan 24, 2014 19:47
tie in circuits into resolution how?
Sasho Nikolov
Jan 24, 2014 19:47
i know you are after some research program, you've hinted and stated that many times. but this is a q&a site. if your question is not well defined, you can't expect good answers, and you can't expect upvotes. you can still pursue your program and ask questions when you have something specific to ask.
Sasho Nikolov
Jan 24, 2014 19:45
bold statement. maybe you are right, but 'give me CC papers that mention the keywords "dag" and "tree"' is not a specific enough question i think
Sasho Nikolov
Jan 24, 2014 19:44
if you ask for circuits vs formulas, we've discussed this in a previous question you had
Sasho Nikolov
Jan 24, 2014 19:43
what's open?