Neural Network "DEEP LEARNING" digit recognition

http://imgur.com/a/7zJml

Stop the presses some retard fucked up their MNIST model.

>>2
what

Oh, look, it's yet another guided brute force multi-objective optimizer that will attract Mentishit posts.

anyone else shocked that it's 2016 and computers still can't read?

"derp learning"

>>6
*sends back to reddit*

What does mentifex think of digit checking

recognize my anus

>>9
Looks like shit, captain.

_{(this space left intentionally unrecognized)}

>>11
good dub

>>12
fuck off

>>13
kill yourself

>>14
only if I can take >>12 out in the explosion.

Deep Gradient-Network Mutation ?

Deep Anus Rotation?

>>7
i guess you weren't around when reddit wasn't a thing yet, but i forgive you.

>>4
Machine learning has had far too marked and rapid a progression in recent years to dismiss off-hand anymore. New techniques such as attention [1, 2], generative-adversarial networks [3], deep q-learning [4], etc, though requiring large clusters of GPUs to train effectively, have pushed forward the state of the art in numerous problem domains substantially and offer a clear path for future gains as well. Corporate support from the likes of Baidu, Google, and Facebook attracts a steady flow of grad students ensuring continued growth and innovation. Naysayers like Sussman will have to eat their words eventually or face complete and utter irrelevancy.

[1] https://arxiv.org/abs/1409.0473
[2] https://arxiv.org/abs/1502.03044
[3] https://arxiv.org/abs/1511.06434
[4] https://arxiv.org/abs/1312.5602

>>19
The rate of progression is as meaningless as it was back in the pre-AI winter days, when logical algorithms had a hugely "marked and rapid a progression" as well. Machine learning will hit the same wall and be relegated to less-sexy practical mechanisms just like logical programming is nowadays. It's just statistics, after all.

>>19
You can't deny this is nothing more than finding maxima with fancy methods though. It's damn good at solving particular problems, sure, but LE SINGULARITY IS NEAR XD retards have no reason to push their shit when the most we can do is crunch numbers to solve overly specific tasks.

If there's one thing I can't deny myself from, it's these dubs

>>21
I think tasks are getting gradually more general but yeah, I don't see either an AI techno-utopia or a world nuked by Skynet in foreseeable future. people believe that because it sounds cool but unfrotunately the truth is more mundane.

>>19
this is just marketing crap, in reality ML is garbage and still can't tell the 3 from the letter 5