/prog/ - Asking for some neuronal power

Name: HAL-9000 2011-04-13 23:59

Hi guyz

Some time ago, a fag posted a thread about evolutionary programming. Well, other fag answered saying "Read Maturana and read Ashby".

I listened to him. Maturana is a cool guy, very interesting ideas, but the Ashby's book got me.

I've only read the 1st chapter. The book is a little dense for me. But, I am a little bored of making SQL queries and fixing bash scripts at work. So, i think it could be an interesting quest to develop a "cybernetic library", so i can learn from the book making programs instead of lame excercises...

So, who want's to help me?

I know perl and C, and also some useless and less interesting stuff. The 1st chapter of Asbhys book is very interesting and lays the basic components, like the transform, the transformation, operands and such.

I mostly need "conceptual" help on how to model such primitives.

Help me!

the book = http://pespmc1.vub.ac.be/books/introcyb.pdf

Name: Anonymous 2011-04-14 0:02

smoke weed everyday

Name: Anonymous 2011-04-14 0:10

>>2
done

Name: Anonymous 2011-04-14 0:14

>>1
For inspiration read books on Real Evolution, how nature selects and propagates genes.

Name: Anonymous 2011-04-14 0:35

>>1
What are your goals? Serious question. I may be able to help.

Name: HAL_9000 2011-04-14 0:57

>>5
Modelling real life problems using the primitives mentioned in the book. i really dont care about speed. im interested in managing complexity, which is the main metric according to ashby.

Name: Anonymous 2011-04-14 3:23

>>6
Oh, sorry I can't help you there.

Name: Anonymous 2011-04-14 5:45

>>1
I can't help you now, but if you come back later (when I read the book) and bump the thread, I'll gladly answer your question.

In the meanwhile, read SICP. You might find your answer, as well as the answers of all of the questions in your life, after you attain satori.

Name: Anonymous 2011-04-14 7:59

>>6
How about some examples.

Name: Anonymous 2011-04-14 10:02

Ok. This is quite simplified, but i hope it reflects the idea

You have a system A, which has a certain amount of "variables". The ammount of these variables is called complexity. Let's call the complexity function C(x).

Then, you have a system B that manages system A. This system has also a certain amount of "variables" (complexity).

I want to study how much complexity is really required on system B to be able to cope with the system A.

Asbhy says that C(B) >= C(A). I want to study how much is REALLY NEEDED.

Why is this needed? Because with this knowledge we can introduce amplificators or reductors of the variety from system A to B or viceversa.

It's a little complicated, but watch this video then it MIGHT make sense.
Is not too long, take a look
http://www.youtube.com/watch?v=BaLHocBdG3A

Name: Anonymous 2011-04-14 10:11

http://krautchan.net/files/1302773399001.jpg

Name: Anonymous 2011-04-14 10:17

>>10
Holy christraping ambiguity.

Name: Anonymous 2011-04-14 10:40

>>12
My post or the video?

Name: Anonymous 2011-04-14 11:09

I couldn't watch that video. Nevermind understand.

Name: Anonymous 2011-04-14 19:03

>>1
>>10
You could try looking at some of the neuroevolution literature on how to develop genetic encoding schemes that minimize the ratio of genetic information to phenotypic information.

Bear with me.

The amount of information that a genetic system can maintain is limited by mutational pressure and by bounds on selective pressure. mutational pressure scales linearly with the length of the genome and has an entropic (destructive) influence on the information in the genome. Meanwhile, selective pressure, which acts to hold mutational pressure at bay, is invariant to genome size and is bounded, over time, by population demographics. As selective pressure increases, survivability in the population goes below levels needed to sustain the size of the population. If such elevated levels of selective pressure persist over a certain period of time, extinction occurs. Thus, genome length must remain small enough that mutational pressure does is not stronger than can be counteracted by sustainable levels of selective pressure.

The relevant implication of this is that in order for a genome to evolve increasingly complex phenotypes, one bit of genetic information must code for increasingly large amounts of phenotypic information. Such encoding is referred to as "implicit encoding." Implicit encoding is apparent in the human brain, which, even in a prenatal state, contains more information than an entire human genome.

Implicit encoding is an ongoing area of research in the area of (artificial) neuroevolution, which is a form of machine learning.

see especially the sections on implicit encoding.

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.147.8975&rep=rep1&type=pdf
http://infoscience.epfl.ch/record/112676/files/FloreanoDuerrMattiussi2008.pdf

Name: Anonymous 2011-04-14 19:28

>>15
A few months ago a friend of mine, who is an up-and-coming nature photographer, decided to spend a day and night alone in the woods outside of our town. She wanted to get photos of the woods and wildlife as naturally as she could for her portfolio. She wasn’t afraid of being alone, as she had camped by herself many times before. She set up a tent in the middle of a small clearing and spent the day taking pictures. She filled up four rolls of film on that trip, but when she went and got them developed she saw four pictures that unsettled her, these four pictures were taken from inside the tent, of her, asleep in the middle of the night.

Name: HAL-9000 2011-04-14 19:39

Well... Is certainly interesting, as is it an application of the definition of information as the difference that makes a difference.

But I'm not interested on modeling mutating components, or transmitting information about the inner state of the elements composing a system. Instead, I'm more interested in the information needed to make the parts cooperate. The system components may learn by themselves or cooperatively, but that's is not my focus.

Is like saying "I'm interested in the messages, not the transmitters".

Of course, the more information the better. But i want to study the complexity of the communication (the possible number of states of a system, which is transmitted to other components of the system), The phenotypic information will certainly reduce the amount of information needed to transmit, but is still creates complexity that has to be considered (amplified or reduced), even if it does change slowly.

I just saw your post a few minutes ago. Now i have a good reading material for the rest of the week. Thanks >>15-san!

:)

Name: Anonymous 2011-04-15 10:23

Hey, 15 here. I just watched the video, and to me he seems to be saying that it should actually be C(B) <= C(A). Or, in other words, the amount of information it takes to control system B is less than the amount of information that fully describes system B's activity. And that sounds right to me. Of course, the lecture is on managerial cybernetics, and so it likely stresses practical constraints on your control system. I suppose maybe its possible, in theory, that achieving perfect control of a system could require greater informational input than the informational complexity of the system. It seems unlikely to me though, since any instruction, in my conception at least, usually leaves some details undefined. The amount of information needed to INFORM an instruction though, is another story.