Artificial Intelligence Stack Exchange is a question and answer site for people interested in conceptual questions about life and challenges in a world where "cognitive" functions can be mimicked in purely digital environment. It only takes a minute to sign up.
Sign up to join this community
What is the status of the capsule networks?
I got an impression that capsule networks turned out not to be so useful in applications more complicated than the MNIST (at least according to this reddit discussion).
Is this really the case? Or can they be a promising research direction (and if so, is there any specific application for which they seem the most promising)?
ML is full with things that supposed to work better (in theory).
Sigmoid function seems better than ReLu.
L1 seems way better than L2.
Spikes neural network seem to be better than standard neural network.
A shallow neural network with a lot of neurons has more parameters than a deep one with the same amount of neurons. So, in theory, has to be more powerful.
Is common to forget the big impact of the training capacity. And easy to get charmed by its theoretical capacity.
Training process are still in diapers. We lack a good understating of it.
Today, Capsules neural network are a very powerful architecture, hard to train.
