AI results

[Skaperen]

when you "train" some artificial intelligence method, what do you end up with when its done with a good set of data?

[Selur]

From what I understand (not much): Assuming you are refering to training a neural network (https://towardsdatascience.com/how-do-we-train-neural-networks-edd985562b73) like real-esrgan (https://github.com/xinntao/Real-ESRGAN/blob/master/Training.md) you end up with a model. Whether that generated is useful or not can only be evaluated by applying it to acctual content.

[Skaperen]

i'm curious what kind of file(s) make up the model. is it programming code in some language? LISP maybe? a bunch of numbers? binary?

[Selur]

You can download ESRGAN models over at https://upscale.wiki/wiki/Model_Database the models usually contains training data for the network.
So it's usually a bunch of headers and lots of binary data containing images (for example in https://data.vision.ee.ethz.ch/cvl/DIV2K/ format which is often used for video/image based neural networks).
You might acctually want to read a bit, https://github.com/krasserm/super-resolution/blob/master/README.md#getting-started has some good links to follow.

[Skaperen]

suppose you wanted to create a model/method/whatever to be able to look at a picture and determine the number of people in it. you have a million example images with the correct number for each, 0 to 10. what steps would be involved to be able to start training? what else would you need to do to have a program you can run that takes an image in a file and print the probably correct number? if you want to "install" this on another computer, how much would need to be copied to the other computer? would it need those million example images?

[Selur]

Since it does not seem that you actually know how machine learning and neural network work you might want to first understand the basics.
About your counting people example: https://www.rapyd.ai/use-case/counting-people-in-images

That's it from me, may be someone else has any motivation to explain all this in details.

Cu Selur



´

[Skaperen]

no, i do not know anything about AI. i don't really care to learn it. what i do want to know about it is the broad picture of what people are doing with it. i have looked at some things that seem to try to teach it, but i just end up with more questions. my background is more along the lines of physics, electronics, computers, system programming, and such.

[Skaperen]

at this point i'm curious how the goal (count people, not animals) gets learned, and what is the final form. are we talking about programmed logic? does coding (such as how to get pixels out of image files) need to be involved? what is the range of what deep learning can do? can this learn to decode the bits that make up a Linux kernel on an ARM platform?

[Skaperen]

what if someone wanted to have AI trained to count people, and not be fooled by a bunch of cats in the picture, and didn't have a bunch of pictures to train on or only had pictures without cats?