How do I train neural network to find me porn that caters to my tastes?
>>59764299
You forgot the greater-than sign in the beginning.
First need data on your past viewing history. Use features like word embedding for tags or fetishes. Combine with maybe convolutional neural nets to extract visual features from vidoes.
You just can't enough raw data. Then you need data like amount watched and need to somehow understand what the sweet spot is..
If it closes too soon then maybe you didn't like it
If it doesn't close for a long time maybe its not "hot"
Etc...
>>59764330
>extract visual features from videos
Liberally impossible
NSA, help me out! Use your powers for good.
>>59764299
That's what tags are for
Firstly, you need to choose a neural network.
I would suggest you do a feed forward net for classification (ie searching up tags), then a deep convolution net with sped up porn to filter out the unwanted variations of porn.
Then I would suggest you watch porn for couple of weeks to train the net.
Then run it forever and it will find what you want.
Hope I helped.
Find new porn website, go view all the model, none fit your taste
>>59764445
Tags just dont work well, people put everything on tag now, not to mention some just plain wrong tag
>>59764988
This. People also duplicate post vids and makeup titles to turn normal porn into incest for example...
make a browser plugin that links to a website where people can make an account, and tag the videos differently. you find people with similar tastes and tags become more meaningful when you follow/friend those people
>>59764422
Nope
A video is a series of images
There are "3D" convolutional layers which can take a series of images.
It might work better with some recurrence though. However recurrent convolutional networks isnt a well explored topic atm.