Last week I was traveling to meet my coordinator in another city, well it was just another day of sun and I was kind of sleepy, but something interesting occurred to me.
No, it was not another idea, it was a perception. Maybe one of the simplest ones, and the one that you just ignore after some days. You know, it is simple, so it is also simple to forget. But it may be very interesting and outstanding if took to the limits...
So let me tell what it was... I was trying to solve a artificial intelligence puzzle, just like I do everyday, hehe. But the problem was about an area of optimization (In fact I can prove that optimization and learning are very similar, but I will not do it today, it is boring and most of you guys should also think it is obvious), so I was just thinking in this direction and so on. Trying to figure out how to improve the minimal functions that are supposed to deal with creating a model of dependencies of variables. Ok I should have lost you know, but do not care about the exact meaning of those words, try to think what it represents for you. Something like "how does it feel" instead of "how does it work".
But in a minute I noticed that a minimal function of the optimization algorithm was a clustering algorithm, ok. No, wait! Clustering of pixels of images? No, of variables.
Ah, well seems the same... and it is. But wait... I was talking about optimization, and that lead me to use dependency of variables , what on the other hand need to have a clustering algorithm, what we can guess needs a... ? optimization? no... deliciously yes! =D
A recursive solution, nice, but this goes much far away, much more than I can understand now. There are places on social sciences talking about clustering and interpreting probabilistic models. Sometimes I am talking with a guy and I just realize that everything we were talking could be transferred in some simple logic puzzles. I know someday I will look at this post and dislike my ingenuity, but I will post this here anyway.
How far away it goes? Well, phylogenetic networks try to create groups that are minimally distant, what does it mean? Clustering! Learning is just an optimization of the minimum of errors. And it goes on my friend, not just with these algorithms, but with others. We are using things here, others are using things there. We are not related, but the things we use are.
What I mean is that we might be confronting, not only with a big theory of everything of the physics, we might have also a big theory of everything of the logic world.
And this logic world could be composed of simpler functions that builds everything. If that is not the case, at least it would help us a lot having a better system of integrating logic in science than the article's system do. You know, it might just be the case that we are stucked in a local optimum of development. Nah, too drastic. With all these people around we might get enough samples and we are developing faster than ever! But better?
What does I mean, well we might all be thinking that we are making better solutions, but we are in fact just limited by a interpretation of our problems. And the things that we solve today could just never be solved at all in these space of solutions of the functions we are using. Yes I am talking about Space of functions, not solutions. Reminds you the computational complexity theory? But I am talking about other similarities of the functions in logical terms not their complexity.
Distance between functions? Well... yes! Maybe it is not possible, maybe it is... who knows? Wait I think I know how... shit this is gonna be difficult to justify...
See you in the next monthly post (it seems to be following this rhythm really strict, hehe)
No comments:
Post a Comment