/ML&DL/ - Machine Learning & Deep Learning General

>>57797910
Let me contribute:
https://github.com/ryanjay0/miles-deep
A deep neural net porn classifier

>>57798220
Also this list is good
https://github.com/jtoy/awesome-tensorflow

>>57798254
Forgot about this too:
https://github.com/owainlewis/awesome-artificial-intelligence

>>57798270
>https://github.com/owainlewis/awesome-artificial-intelligence
>>57798254

These are great, thanks!

Do really deep/ recurrent networks of ReLU or maxout neurons tend to suffer from divergence? In other words, will the values just keep getting bigger and bigger as you go through the network, to the point where they would overflow if the network is deep enough?

I want to use straight ReLU or maxout neurons in an RNN without all that LSTM bullshit

>>57798983
I don't really know what you mean, but why couldnt you renormalize them?

An Artificial Neural Network is unique per input, right?
So if I'm doing an logical XOR training, there should be a separate network for input {0,0}, {0, 1}, etc
And if I want to test my AI for a specific input, I will call a network depending on my input? Is my understanding correct?

>>57800809
>An Artificial Neural Network is unique per input, right?


what

What does the OP picture have to do with machine learning? Could you use something appropriate next time please?

>>57797910
how much programming do i need to know to start? where do i get data to use?

i know a little bit of python

>>57801132
You don't need to know much beyond the basics of Python then learn as you need. For data look into Kaggle

>>57800993
Nothing. This thread will 404 faster than you can blink if you don't do use something that grabs attention.

>>57797910
I busy trying to use DIGITs hosted on Amazon AWS, I'm not a progammer or anything.

I just follow instructions.

I have had to learn many things but I think it is worth it.

I want to classify the content of images using a trained model from the imagenet dataset using cafe, I think the one with 1000 categories. Then remove the last layer and retrain on my dataset.

Like the guy with cucumbers but on a server, using gpu and graphical interface of DIGITS.

I would run it locally but I have to use this machine for other work that require windows.

Docker dose not work nor does other virtualization options, something wrong with my environment.

Thought the cloud options would be faster I was wrong maybe.

>>57803356
> busy trying to use DIGITs hosted on Amazon AWS, I'm not a progammer or anything.

Can you hook me up with the tutorials you used? I know nothing about this but interested.

How much work on your end did you do, or is this all pre-written for you? I'm interested in more hands on implementation and using the cloud to do the computation stuff.

Also, how much does it cost? I'm a broke student

>>57799173
Wouldn't messing with the activations like that fuck with the gradients and cause it to mislearn?

>>57798983
>without all that lstm bullshit
hahahaha ok

>>57803404
dunno about digits but you get a bunch of free stuff from github and their student package including vps money to host your own projects

>>57803461
>implying LSTM's aren't a dodgy hack

>>57804738
>implying taylor approximations aren't a dodgy hack

>>57804768
>implying that's not a non sequitur

File: obwn.jpg (210KB, 1920x1440px) Image search: [Google]

210KB, 1920x1440px

nth for waifu AI's

>>57804966
THINN

>>57804738
>implying the brain isn't a dodgy hack

I am working on a CONVNET to classify images as anime or not anime. Gonna work it into a chrome addon to block anime images.

just kidding i haven't even started but i'm interested in the idea. anyone wanna start an open source project for this??

>>57807188
run it in a TempleOS vm. that way it will at least know God.