Self-Attention Agreement Among CapsulesSelf-Attention Agreement Among Capsules • September 30th, 2021
• From entity to classification: The network computes the probability of presence of an entity in the image. The prediction represents the classification of the image.
Self-Attention Agreement Among CapsulesSelf-Attention Agreement Among Capsules • October 11th, 2021
The output of the capsule vectors represent the probability of an object of been present in the given input or not. These output vectors can exceed value one, depending on the out- put, so to make the output vector represent a probability, a non linear squashing function is used to restrict the vec- tor length to 1, where sj is input to capsule j and vj is the output.
