A Branching and Merging Convolutional Network with Homogeneous Filter Capsules

24 Jan 2020 Adam Byerly Tatiana Kalganova Ian Dear

We present a convolutional neural network design with additional branches after certain convolutions so that we can extract features with differing effective receptive fields and levels of abstraction. From each branch, we transform each of the final filters into a pair of homogeneous vector capsules... (read more)

PDF Abstract

Datasets


Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification MNIST Branching/Merging CNN + Homogeneous Filter Capsules Percentage error 0.16 # 1
Accuracy 99.84 # 1
Trainable Parameters 1,514,187 # 1

Methods used in the Paper


METHOD TYPE
Adam
Stochastic Optimization