Can someone pls explain machine learning

It's statistics done by people who don't understand statistics.

If your friends jump off a bridge, you jump too.

pattern recognition and curve fitting meet SJW

it's when you windows 10 machine learns about you

Gimmick word for data collection done by computers.

no u

it's a fucking meme

there is no "learning", it's just algorithms made by humans and then the CPU just does brute force number crunching

it's a dumb buzzword, same as "AI" or other such bullshit

Statistics, heuristics, matrix calculations, layers.

You create a "program randomizer" or sorts that generates a random code, then you create a program that gives the result of the other program a score.
Then you keep changing the random program to increase the score more and more until it does what you want it to do.

An algo that uses brute force methods to find solution to a specific state.

State means a goal, a problem, an environment, etc.

If the state changes even a tiny bit, then you need to brute force a solution again and again.

Machine Learning is only good for very specific goals and problems. It does not work in the real messy world with unlimited states.

In other words, machine learning does not scale! This was already concluded by the AI researchers 40 years ago.

What does scale?

brilliant.org/courses/artificial-neural-networks/

brilliant.org/courses/machine-learning/

Attached: pepo autismo beated with a hammer.gif (487x560, 898K)

It's a bunch of if statememts

Attached: 1528811151167.png (570x960, 712K)

Statistics wrapped in basic algorithms to make it seem very sexy.

I think the easiest solution is to create an open-source common sense library/API and let the people contribute to this project and watch it evolve.

The main problem is capitalism and the fact that humans are selfish and greedy by default.

Big tech companies hire potential legendary geniuses to become dummy. They mostly hire them so that they will not go to other companies or do revolutionary shits that will disrupt their business.

Most algorithms are already out there for anyone to implement

And what you're talking about has nothing to do with scalability

here you go user

Attached: 35142545_10155854322778040_6623103573481750528_n.jpg (371x439, 29K)

...

It's racist and should be illegal

> Most algorithms are already out there for anyone to implement
Almost all algorithms we have are already been there since the 80s.

Usually you don't have convex functions

diffy q and linear algebra team up to gangbang stats

homes.cs.washington.edu/~pedrod/papers/cacm12.pdf

Isn't it just trying to get something random to do it in such a way that it's as close to 1 as a correlation as possible? So just going through millions of times getting closer from 0.05 to 0.4 to 0.9 etc? So just statistics and shit?

Applied statistics. Numerical methods for finding maximum likelihood estimators and decisions boundaries.
You need to be above brainlet level to understand the underlying mathematical foundations properly, but you can just hack it and write your own stupid net in a weekend with Keras and feel like you achieved something.

SQL

that doesn't stop you from using the same algorithms

Attached: 1528752378286.jpg (1068x1085, 158K)