In a way, capsule representation seems to have an overlap with Grossberg's scaling and orientation subsystems of ART2 network. These problems are also fundamental to all graphic systems. Also, the binding problem discussed by Malsburg (1990) with an example of visual cortex, where he talks about the need higher levels of abstraction while the lower level recognition is confined to small parts of receptive field. It looks like we are going in circles, rather spirals of technological innovation. But this time we have hardware supporting the theory. Long live AI!
In a way, capsule representation seems to have an overlap with Grossberg's scaling and orientation subsystems of ART2 network. These problems are also fundamental to all graphic systems. Also, the binding problem discussed by Malsburg (1990) with an example of visual cortex, where he talks about the need higher levels of abstraction while the lower level recognition is confined to small parts of receptive field. It looks like we are going in circles, rather spirals of technological innovation. But this time we have hardware supporting the theory. Long live AI!
It is hard to understand, but was very interesting.
I've watched this video for 3 times already and now started to understand some ideas.
the vid is record in 2014, only after 4 yrs, this vid appear in 2018, just wonder
Interesting!
I have half a mind to create an ML model just to detect whenever he says "um" and create a supercut video of all those instances.
Hinton年轻的时候还是一个大帅哥的，现在已成为教父了，岁月啊