Deep fucking dreaming?

>>8562807
https://www.youtube.com/watch?v=BsSmBPmPeYQ

>>8562824
That guy looks like a cuck. Fuck you, you dumb fucking nerds.

Why does it see dogs in everything?

Fuck you.

when i kinna take LSD this kinda stuff i see

like, xactly like dis

so you tellin me

computers on LSD too now? world far gone man, wold fare gon.... maybe AI wont dstroy mankind after all.... red pilled n all

>>8562807
A neural network is basically a bunch of functions mangled together with a lot of parameters. To train the network you gradually alter all those parameters to move them closer to the optimum (which is the minimum of a loss function, which is just the error of the classification in the simplest case). This is done in countless tiny steps until you get whatever you want. The idea is that you have so many parameters in your classifier that they will eventually cover the whole information needed to "understand" the classes.

Deep dreaming is now a somewhat reverse process. You give the trained network some image and a class you want (here: dogs). Then you ask the network how much dog it sees. After that you alter the pixels of the image in a direction to increase the "dogness". You do this over and over and eventually you end up with an image like above.

>>8562844
Dude I just used some filters of a deep dream website and that's what I got maaan?

>>8562848
Ok

>>8562844
>increase the dogness

for some reason i find this hilarious

File: dream_ec4e758478.jpg (196KB, 600x600px) Image search: [Google]

196KB, 600x600px

>>8562844
Can I set it up to increase the "Pepeness" of an image?

>>8562807
The computer likes dogs?

I don't think it likes dogs that much. Out of all the possible squiggles you could focus on in that image, you and i go first for the ones that look like a face, and don't pay attention to the other parts because we don't recognize anything in them. iirc deepdream tunes the input image so that the net actually does see something in those otherwise unrecognizable parts of the image

can any mlfags confirm whether it might be due to some kind of imbalance in training data?

>>8562807
Its programed to find dogs

>>8562882
Somewhat... But it's still relying on matching features from images that belong to non-pepe classes.

>>8562844
>A neural network is basically a bunch of functions mangled together with a lot of parameters.
That's really just one function.

>>8562807
There's already a computerphile video on this you fucking faggot

>>8562807
pretty sure i've inadvertently lucid dreamed before on many occasions. that being said, inadvertently is basically not a lucid dream at all. though often times i've managed it with feasible measure of consciously attainable success even if i'm not an active practitioner..
>>8562844
yep, lucid dreaming is at base just specifying parameters for dreaming within the parameters of experiencing such things within and slightly beyond known possibilities of experience. outside of that is no longer a dreaming experience within control and quickly becomes a nightmare/transcendent experience. one emboldens you with valuable knowledge while the other drives you further into madness though i'm sure intelligent individuals are adequately able to handle neurological input beyond normal capacity.
both will afford you insight into what can be possible but one belies that which will take you a step beyond the capacity to experience and bility manipulate the present.
that being said, madness is a feasible cost to afford for the possibility of greatness if you at all care about the merits of the physical world. those who don't care for such things, well, absolute lucid experience is a recommendable experience.

>>8563334
*ability to

>>8563334
>>8563335
i've realised this is not what this thread is about i'm very drunk merry christmas my dudes
saged my fucking post

>>8563272
You don't know shit, shut the fuck up

>>8562807
Corgi Detection Algorithm