Capsule Networks Explained – learn more | reddit discussion
Using a novel architecture that mimics the human vision system, Capsule Networks strives for translation equivariance instead of translation invariance, allowing it to generalize to a greater degree from different view points with less training data.
You can check out a barebone implementation of Capsule Network here, which is just a cleaned up version of gram.ai’s implementaion.
- CNN: Translation Equivariance and Invariance
- The Impact of Imbalanced Training Data for Convolutional Neural Networks
- What is wrong with Convolutional Neural Networks?
- Dynamic Routing Between Capsules