Capsule neural networks

A system comprising:

• one or more computers;
• and
• one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:
• receiving a network input to a neural network;
• processing the network input using the neural network to generate a network output, wherein:
• the neural network includes:
• (i) a first capsule layer comprising a plurality of first capsules, and (ii) a second capsule layer comprising a plurality of second capsules;
• and
• processing the network input using the neural network comprises:
• determining a layer input to the first capsule layer based on the network input to the neural network;
• processing the layer input to the first capsule layer to generate a layer output of the first capsule layer, wherein the layer output of the first capsule layer comprises a respective first capsule output of each first capsule in the first capsule layer;
• and
• processing the layer output of the first capsule layer using the second capsule layer to generate a layer output of the second capsule layer, comprising:
• iteratively adjusting values of a set of routing factors, over a plurality of update iterations, based on the layer output of the first capsule layer to determine final values of the routing factors,
• wherein the values of the routing factors define a routing of the first capsule outputs of the first capsules in the first capsule layer to corresponding second capsules in the second capsule layer;
• processing the first layer output of the first capsule layer, in accordance with the final values of the routing factors and a set of capsule layer parameters of the second capsule layer, to generate the layer output of the second capsule layer;
• and
• generating the network output of the neural network based on the layer output of the second capsule layer.

One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

• receiving a network input to a neural network;
• processing the network input using the neural network to generate a network output, wherein:
• the neural network includes:
• (i) a first capsule layer comprising a plurality of first capsules, and (ii) a second capsule layer comprising a plurality of second capsules;
• and
• processing the network input using the neural network comprises:
• determining a layer input to the first capsule layer based on the network input to the neural network;
• processing the layer input to the first capsule layer to generate a layer output of the first capsule layer, wherein the layer output of the first capsule layer comprises a respective first capsule output of each first capsule in the first capsule layer;
• and
• processing the layer output of the first capsule layer using the second capsule layer to generate a layer output of the second capsule layer, comprising:
• iteratively adjusting values of a set of routing factors, over a plurality of update iterations, based on the layer output of the first capsule layer to determine final values of the routing factors,
• wherein the values of the routing factors define a routing of the first capsule outputs of the first capsules in the first capsule layer to corresponding second capsules in the second capsule layer;
• processing the first layer output of the first capsule layer, in accordance with the final values of the routing factors and a set of capsule layer parameters of the second capsule layer, to generate the layer output of the second capsule layer;
• and
• generating the network output of the neural network based on the layer output of the second capsule layer.