Deep Learning thread
What is anon working on?
it's a meme
just neural nets that require a so much training (human effort) it'll never be cost effective
>>57582256
>it's a meme
that's the real meme
AI for my future waifu(bot). All the current implementations of "neural networks" are however more geared towards solving highly non-linear problems by throwing enough data and tweaking at it that the algo starts producing good enough answers, so I had start from scratch. Progress is slow.
>>57582881
>Progress is slow.
Empty .c file, amirite?
trying to understand gradient descent
>>57582256
>it will never be cost effective
U wot m8
Modern GPU can do deep learning pretty effectively. The most advanced Go AI, AlphaGo, only took 50 GPU and 3 weeks training for it to beat the best human players. Since its scalable, you can scale it down or up and it will go either way. Once its trained, it can be run on laptop gpu.
Let's build a dataset of pairs of images and pepes of those images.
Then train a model to create new pepes from new images
>>57583047
I can already predict that you'd end up with warped pepe faces everywhere on the produced images. They'd fit, superficially, but they'd be trippy as shit, like with that deep dream stuff.
Trying to train a neural network to tell me a good introductory book on neural networks, deep learning and meme science.
>>57582914
Coding is not the hard part, senpai. It's only laborious, I might even outsource some of it to pajeets, once I get the sw architecture and design fleshed out
I am still starting with deep learning. Thought is was a meme because of the fancy name and all this talk about how this will produce "real AI".
But as >>57582881 said: It is more about optimization of non-linear functions. "Deep learning" is just the marketable name. And the more I read about it the more I think it not just a fleeting trend.
Just finished with the basics of Machine Learning from Coursera. Made a nice PDF from www.deeplearningbook.org for reading in the next few months and I am trying to work with the OpenCL branch of Caffe because I am a dumb fuck and have an AMD GPU.
For starters I thought I can use some labeled pictures from a Booru for some simple classification problems. And I should really start reading some papers…
>>57583092
kek
was trying to make hologram system but 90% of the concept got stolen by a fagg from kickstarter
Working on a space heater
>>57582256
>He doesn't use MTurk for generating his training data
3rd year CS student. Love the algorithm/data structure side of CS, not that muh startup muh html shit
what's the best way to dip my toes in this field? I'm thinking grad school is in the cards, but I'd like experience with something in the machine learning/ai/meme science field before making the leap