Using time as a supervision mechanism for self-supervised learning:https://arxiv.org/abs/1605.08104
-
Using spatial relationships: http://arxiv.org/abs/1603.09246
It is likely that humans rely on both temporal continuity and spatial coherence to learn about visual features. So why not combine both?
-
-
One thing I like for this is 'for data set f(x), predict x given f' (both x, f high dim. vectors). Learn a map of the data.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This was a cool paper on the same theme too: https://arxiv.org/abs/1511.06811 "Learning visual groups from co-occurrences in space and time"
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
aren't those captured in the 'intuitive physics'/'reverse game engine's models of learning, for example http://dspace.mit.edu/handle/1721.1/89115 … ?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I thought there was some work that did this but it was background subtraction.. not quite the same http://link.springer.com/chapter/10.1007/978-3-642-33783-3_17 …
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.