I'm not sure about 5G.

Also, the image recognition is NOT lame.

It can recognize 22,000 image classes.

You go train a model like that on your home computer.

Using someone else's pre made code is

can it recognize khazar milkers tho

Ah, but poo-pooing someone else's work is greaT?

5g is good but it's entirely Chinese controlled as if now

Oh my god i knew it

You are supposed to re-use models.

That's the point of having it and training it for months.

@Tero Is a trap indeed

I don't have a datacenter in my backyard.

@Deleted User talking about tensor flow not the model lolo

I can train a neural network, no problem. There is nothing difficult about it.

Ah.

So using Tensorflow is lame. XD

Gotcha.

<:GWjiangoOmegaLUL:389904150886088723>

Half of the AI projects are lame then.

And maybe they are...

But Tensforflow is still cool.

Actually, Keras is better.

Or easier at least.

I trained a neural network with just three layers to an accuracy of 96% on the MNIST dataset.

I think it took 5 minutes on my laptop.

I also put peanut butter on my nose.

/thread

a neural network is basically a self-improving AI?

@Deleted User

It's like a structure for statistical learning.

You have weights that determine how strong each neuron contributes.

(((statistical learning))) what the hell

>Whoa, no one said these Central Americans were going to live in MY city!

Each neuron is basically a function of its inputs from the previous layer.

Then you look at how far your output is from the intended result.

women literally unable to think in general principles, only about what is right in front of them, the image

You can then use gradient descent or another error estimation method to back propagate the changes through your neural network.

wow

Common activation functions are sigmoid, tanh, RELU, identity function, and extended sigmoid.

i gotta get in this shit one day, to busy rn tho

That's the neat part: it's easy with tensforflow.