Humans think in terms of sequences of actions. Arguably the best way to emulate human creative behaviors is by generating data sequentially.
-
This stands in sharp contrast with the current focus of the creative AI community, which is all about jointly generating a grid of pixels.
Humans don't know about pixels. They draw one brush stroke at a time, sequentially, incrementally.
-
-
But getting this kind of training data is harder, right?
- Show replies
New conversation -
-
-
Luckily some smart people (read:
@hardmaru) have been paving the roads towards this https://github.com/hardmaru/sketch-rnn … https://github.com/tensorflow/magenta/blob/master/magenta/models/sketch_rnn/ …Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
My personal experience is that ideas can come all at once as a picture or an object
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Berkeley in 1709: “consider the case of an intelligence which [can] see perfectly well but [has] no sense of touch”
-
“men are tempted to think that flat or plane figures are immediate objects of sight, though they acknowledge solids are not.”
- Show replies
New conversation -
-
-
This seems close to Jeff Hawkins' line of thinking, doesn't it?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
But what about nature? I am primarily interested in generating realistic / photorealistic images.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.