On imagenet, it can tell the difference between a school bus and a cat. For semantic segmentation, it can find all pedestrians, buildings, cars, etc. in an image. These are brittle, but if it didn't have them, we could say it lacked even this minimal common sense.
agree there is not going to be a single magic moment; but what common sense would you attribute to current deep learning?https://twitter.com/tdietterich/status/960967687349284865 …
-
-
-
I think there is no common understanding of what common sense is. It actually is an entire bag of things. For vision that could be the ability to attribute shadows to sources of illumination and objects that cast them. And a million other things, that we cannot yet do.
- 2 more replies
New conversation -
-
-
The way that we build a system should not affect our evaluation of whether it knows a certain fact or can perform a certain capability. Obviously as engineers we prefer easier ways, but if a method works, it works.
-
Sorry to have deleted previous tweet in order to correct its English. "Common sense is not about identifying a bus after one has seen a million busses before. Common sense is identifying a bus after somebody has described a bus in English with may be a picture or two. "
End of conversation
New conversation -
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.