only a tiny subset of these valid images would constitute a "dense sampling" of the space of natural images. The same is true of many other spaces, including the space of Go board positions.
Show this thread
Just because your representation of choice for a problem space is high-dimensional doesn't mean that the problem is complex. There are impossibly many possible 400x400x3 uint8 tensors, but only a small subset of them are interpretable as natural images, and
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
True, but estimating domain complexity is very hard. And domain experts' pay is strongly correlated with the perceived complexity of their domain, creating bad incentives
-
This is such a deep deep point about "expertise." After working for almost thirty years I've finally realized that people are paid based on how effectively they obfuscate and misrepresent their understanding of the domain of their field.
End of conversation
New conversation -
-
-
Although I have always wondered how much of the manifold complexity of P(x) is dependent on the image capturing sensors. They are quite handcrafty with lots of parameters tuned by people for visually pleasing represention. Thoughts?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Hi, I just buy your book today. Cheer,
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.