So neural networks can look at a real picture and categorize for you, like Google Photo or facebook. They can find people’s faces and automatically tag them. They can read hand writings and analyze signatures. Self-driving cars use neural network to learn how to drive. Neural networks can even look at a picture and describe what’s in it in correct English.
Now it can even do abstract recognition, like a doodle…
I remember having an argument with my psychology professor (25 years ago) about self-driving cars. My assertion was that neural nets was the way to go, thus allowing the car to figure out what constitutes ‘good driving’. He was all for a rule-based approach and pretty much called me an idiot in front of our tutorial group.
This was back in the days when training a feedforward 50-node neural net took several minutes on a Sun workstation, so I suppose I can excuse him for being an idiot.
Even 5 years ago, lots of paper submissions were rejected just for using neural net as the way to solve the problem. Things didn’t really turn around until 2011, when GPU powered mixture of convolutional layers, max-pooling layers, and classification layers began to out perform all other methods in image recognition competitions.
Back when yahoo had their online pictionary game, there’s always gonna be a guy trying to draw male genitalia for everything, and usually that guy gets booted from the game room.
some Taiwanese guy tried this on google, and google played on like a champ…
Why are men compelled to do that? I mean there are cave paintings that look like the doodles in that link.
I just tried it, and it never guesses you’ve drawn a huge knob instead of the Eiffel tower or a cannon.
Actually it’s a bit of a cheat, because it has a priori knowledge of what the image is supposed to be (‘draw a … in 20 seconds’) and simply has to guess a yes/no match. As opposed to actually guessing what you’ve drawn. Not much of a technical challenge.
No, it doesn´t know what you are asked to draw, I think. But what it does know is the list of possible things you could be painting, and that eases up the guessing a lot.
This reminds me of something much older but really cool IMO:
It’s not working.
Continuing the general theme of puerile humour:
[quote]Q18. Get your mind out of the gutter! I am not allowed to talk about stuff like this, but, I am guessing that it has something to do with sex?
Right, Wrong, Close
Does it hold water? No.
Is it white? Sometimes.
Can you sit on it? Yes.
Does it get really hot? Maybe.
Would you find it on a farm? Yes.
Is it larger than a pound of butter? Depends.
Is it straight? Yes.
Does it open? No.
Do you use it in public? Rarely.
Can it make a sound? No.
Does it bring joy to people? Yes.
Does it have scales? Doubtful.
Can you find it in a church? Probably.
Does it help accomplish tasks? Yes.
Does it have a spine? No.
Does it provide protection? No.
Is it usually warm? Yes.
It is classified as Unknown.[/quote]
It did pretty well, I thought. Very clever that it can cope with ‘soft’ decision processes (ie., answers other than yes/no).
Depends on the person 
Indeed. I like that too.
This thing has been around for a while, and IMO is very cool, cooler than the google doodle guesser thing in a way.