Author Topic: Graph Isomorphism in Quasipolynomial Time  (Read 600 times)

0 Members and 1 Guest are viewing this topic.

Offline Kafein

  • King
  • **********
  • Renown: 2203
  • Infamy: 808
  • cRPG Player Sir White Rook A Gentleman and a Scholar
    • View Profile
Graph Isomorphism in Quasipolynomial Time
« on: December 18, 2015, 11:27:47 pm »
+3
Forget about shootings, this is the news of the decade.


Offline Leshma

  • Kickstarter Addict
  • King
  • **********
  • Renown: 590
  • Infamy: 389
  • cRPG Player Sir White Rook A Gentleman and a Scholar
  • VOTE 2024
    • View Profile
Re: Graph Isomorphism in Quasipolynomial Time
« Reply #1 on: December 19, 2015, 12:53:54 am »
0
Does this mean they figured out one physics theory to rule them all, which will allow us to interpret what happens to astronauts who fall into black hole? (assuming they somehow survive tidal forces and extreme heat)

Jokes aside, there was a time when I was in contact with stuff they are mentioning in this video but that time is long gone. Due to my line of work not requiring any of this, I forgot almost everything yelp. Probably has something to do with me barely passing final exam and this was by far most complicated shit they tried to teach us at university.

As someone who obviously have strong ties to university and is probably fluent in this shit, could you spend some of your free time trying to explain to simpletons what's this all about? You know what Einstein said, if you can't explain it in a simple way, then you don't have good knowledge of that subject.

Edit: There's something I remember. Most of us took this shit literally, trying to learn it as some kind of process. There were things you had to do in certain order, pretty much like solving most mathematical problems with formulas. Know you had quite a few options to solve some issues that would pop up in the process, because you could end up with some shit math couldn't solve. Don't know if this make any sense to you, but it seems to me they found out a way how to solve some of those equations math couldn't solve previously. Is that even partially correct?
« Last Edit: December 19, 2015, 12:59:26 am by Leshma »

Offline Molly

  • King
  • **********
  • Renown: 1860
  • Infamy: 693
  • cRPG Player Sir Black Rook A Gentleman and a Scholar
    • View Profile
    • For the glorious Khorin...
  • Game nicks: Molly
Re: Graph Isomorphism in Quasipolynomial Time
« Reply #2 on: December 19, 2015, 11:02:59 am »
0
Watched something like 20 min, got a huge flashback to my math courses, skipped to the end with the questions...

Another similarity to my math courses: this feeling of having no idea what just went down :lol:


I hate this kind of highly theoretical math!  :oops:
When west germany annexed east germany, nobody moved a finger too.

Offline Moncho

  • King
  • **********
  • Renown: 1127
  • Infamy: 221
  • cRPG Player Sir Black Bishop A Gentleman and a Scholar
    • View Profile
  • Game nicks: Moncho, Some_Random_STF, Some_Random_Troll
Re: Graph Isomorphism in Quasipolynomial Time
« Reply #3 on: December 19, 2015, 11:44:39 am »
+2
Let me try to give some info. Hope I didn't ramble too much and that is followable, this article exposes it better than I do: http://news.sciencemag.org/math/2015/11/mathematician-claims-breakthrough-complexity-theory .

The graph isomorphism problem, which is that given any two finite graphs you can say whether they are the "same" (up to a relabelling of the edges), that is called isomorphic.
(click to show/hide)
They may look different but they are essentially equal, and share all properties. Eg the two below are isomorphic:
(click to show/hide)

This problem is one of a few that is not known to be solvable in polynomial time nor NP-complete, which is almost certainly the most important open problem in complexity. P type problems are considered "easy", whereas NP-complete are at least as hard as any other NP problem, thus some of the hardest out there.

And now there is this algorithm by Babai which reduces it to quasi polynomial time, that is a much quicker than previous algorithms in the sense that it is almost polynomial in time, thus brings the complexity of the problem much closer to polynomial time (exp(polylog(n))). As such, it places this problem in an area of the problem much closer to P, whereas it was thought it was NP.

For applications, the size of networks and computational speed is probably to the point that this is not too big of an improvement, as the "slower" algorithms still tend to get it done quickly enough. Besides, knowing whether two networks are the same is not usually the question, although some others reduce to this one. But often it is only after the result comes in that new ways to use it are thought of (often coming from the ideas used in the proof of the statement itself).

Offline Kafein

  • King
  • **********
  • Renown: 2203
  • Infamy: 808
  • cRPG Player Sir White Rook A Gentleman and a Scholar
    • View Profile
Re: Graph Isomorphism in Quasipolynomial Time
« Reply #4 on: December 20, 2015, 01:37:56 pm »
+1
He uses a lot of previous results that would take hours to demonstrate. Usually in this kind of talk if you're not comfortable with the literature you can quickly get lost.

But basically the scheme is to look at local areas of some structure built on the graph, determine of which "type" each of these local areas are, and infer an efficient reduction of the problem from that knowledge. The whole method is "dirty" because it depends on the uninformed choice of these areas and more importantly it branches using completely different methods depending on the structure of the input. It has quasi-polynomial complexity merely because we don't know of any input that is more difficult to solve than that.