/sci/ - Science & Math

>>10874225

>How do I into chaos theory?

learn about differential equations

>>10874225
Read pic related

>Even if we make the wildly improbable assumption that the behavior of our society could be predicted through the manipulation of, say, a million trillion simultaneous equations and that sufficient computing power to conduct such manipulation were available, collection of the data necessary for insertion of the appropriate numbers into the equations would be impracticable, especially since the data would have to meet impossibly high standards of precision if the predictions were expected to remain valid over any considerable interval of time. Edward Lorenz, a meteorologist, was the first to call widespread attention to the fact that even the most minute inaccuracy in the data provided can totally invalidate a prediction about the behavior of a complex system. This fact came to be called the "butterfly effect" because in 1972, at a meeting of the American Association for the Advancement of Science, Lorenz gave a talk that he titled "Predictability: Does the Flap of a Butterfly's Wings in Brazil Set Off a Tornado in Texas?" Lorenz's work is said to have been the inspiration for the development of what is called "chaos theory" -the butterfly effect being an example of "chaotic" behavior.

>>10874225
Why chaos theory? You know they have actual textbooks over neural network modelling, right? Plus, a lot of the contents of hidden layers aren't really interpretable by humans, so what do you mean by "analyze"?

>>10874243
>never heard of Cambridge Analytica
They used Big Data to influence voters bud, these people can definitely predict you. That's what Facebook and Google use your data for.
>being against AI
So, you want a future with dog shit technology and dog shit medicine? Contain it to yourself. Some of us actually want a decent future.

>>10874259
Technology won't give humanity a decent future, retard.

>But when all people have become useless, self-prop systems will find no advantage in taking care of anyone. The techies themselves insist that machines will soon surpass humans in intelligence. When that happens, people will be superfluous and natural selection will favor systems that eliminate them-if not abruptly, then in a series of stages so that the risk of rebellion will be minimized.

>Even though the technological world-system still needs large numbers of people for the present, there are now more superfluous humans than there have been in the past because technology has replaced people in many jobs and is making inroads even into occupations formerly thought to require human intelligence. Consequently, under the pressure of economic competition, the world's dominant self-prop systems are already allowing a certain degree of callousness to creep into their treatment of superfluous individuals. In the United States and Europe, pensions and other benefits for retired, disabled, unemployed, and other unproductive persons are being substantially reduced; at least in the U. S., poverty is increasing; and these facts may well indicate the general trend of the future, though there will doubtless be ups and downs.

>>10874252
>You know they have actual textbooks over neural network modelling
Yes, but does that cover _why_ neural networks work? Most of the stuff I've studied so far has been how to create solutions which work, but very little has been about why it actually works. And further experiments I've done has shown that the architecture of the models doesn't matter at all as much as the amount of parameters which are being trained.

>Why chaos theory?
Because I'm always amazed how looking at the patterns of a cup of tea which i just poured milk into is basically completely unpredictable. Simple things which becomes insanely complex are interesting. If i can understand/grasp that, or at least understand and predict that (maybe using neural networks/similar machine learning), I might be able to predict all sorts of things. Also: in order to create better neural networks.

>>10874252
>Plus, a lot of the contents of hidden layers aren't really interpretable by humans
Missed this. This is my point though. The latent factors should be interpretable with sufficient understanding in chaos theory.

My theory is that neural networks work because we create our semi-controlled chaos system which's properties we train. But, seeing that I don't know anything about chaos theory, I don't know if this is the case.

>>10874243
lets say we make 7 billions human clone and move out of the solar system. wed have a running simulation of earth at a sensible economic price and computing power.

>>10874271

So you know nothing about chaos theory, yet you think it can explain how a neural network works.

You've never opened a serious statistics or math textbook, have you?

Neural networks aren't some kind of black magic that nobody understands.

If you want to work on something interesting, then develop a theory that rigorously explains how genetic algorithms work instead.

>>10874309
>You've never opened a serious statistics or math textbook, have you?
Nope, that's what I'm asking for.

>Neural networks aren't some kind of black magic that nobody understands.
Where can I find out how they work?

>>10874315

Nonlinear Dynamics and Chaos by Steven H. Strogatz

Pattern recognition and machine learning by Christopher M. Bishop

>>10874318
Thank you, have a webm.

>>10874262
So, you must be posting on 4chan from your Amish farm then?

>>10874225
>How do I into chaos theory?
Write a script to make a animated plot of the Lorenz attractor. Then mess around with the initial condition a little see what it does. Then read about the Lyapunov exponent. After that, you know about a quarter to a half of all of chaos theory.

>>10874266
That's not really Chaos Theory, anon. You seem to be more interested in Statistics and possibly Probability. For neural networks, you can study Bayesian Statistics and Mathematical Probability or even look into Statistical Learning. Also, understanding "why" they work is diving into what computing is in terms of what the brain does exactly (which is a cutting-edge research field), so you're getting deep into Artificial Intelligence and possibly neuroscience.

>>10874225
if you know calculus and some linear algebra, pick up strogatz "nonlinear dynamics and chaos"

>>10874225
>I want to see if i can analyze neural networks using chaos theory stuff
nice bait

>>10874266
>>10874271
you have no idea how retard you sound, hurry up and pick book to understand why are you embarrassing yourself

>>10874271
>My theory is that neural networks work because we create our semi-controlled chaos system which's properties we train.
What you're really interested in then is the line/relationship between chaotic and stochastic systems. There is a relationship of sorts between """chaos theory""" and """machine learning""", but it's a fairly trite one.

>>10874225
it just has a nice sounding name it ins't really all that interesting.

Literally just sensitivity to initial conditions. same initial condition = same result , slightly different initial condition = very different result.

>>10874654
Is sha-256 a chaotic or stochastic system? In the eyes of mathematicians, I mean. If no: is sha-512? Where does the line go?

>>10875819
>same initial condition = same result ,
What makes it interesting to me is that there's infinitely many slightly different conditions. What really matters?

>>10875833
>Is sha-256 a chaotic or stochastic system?
It's a highly chaotic function, yes.

>>10874262
>In the United States and Europe, pensions and other benefits for retired, disabled, unemployed, and other unproductive persons are being substantially reduced
excellent, a return to the natural order

>>10874259
People are way more complex than a computer can predict. Neurologists have yet to even understand the human brain fully.
Furthermore, even the US government has tried brainwashing people using LSD under MKUltra and did not succeed. My point is that humans are not machines that can be told what to do entirely, though they can be suggested on what to do.

>>10874259
Influencing the people who actually bother to vote is about as impressive to influencing brute animals.

>>10875889
neurologists dont study how the brain works, they study the causes of brain illness and how to treat it.

>>10875891
Cambridge Analytica's trick was influencing people who would not normally bother to vote.

>>10874262
>When that happens, people will be superfluous and natural selection will favor systems that eliminate them-if not abruptly, then in a series of stages so that the risk of rebellion will be minimized.

What a prescience.

>>10874225
https://colah.github.io/

>>10875889
You don't need to understand them fully, you just need the person's data relevant to your problem. If the profiling methods you are using for the problem at hand are doing poorly, the learning model will adjust itself and try new approaches until it finds a good profiling method. That is the magic of data science with social engineering and psychology.