Analyzing Spatially-Sparse Data Based on Submanifold Sparse Convolutional Neural Networks

One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

• access a plurality of content objects;
• generate a plurality of voxelized representations for the plurality of content objects, respectively;
• generate, based on one or more sparse convolutions, one or more building blocks, wherein generating each of the one or more building blocks comprises:
• determine, based on the voxelized representation of each of the plurality of content objects, one or more active sites for each of the plurality of content objects;
• and
• apply, to the one or more active sites, the one or more sparse convolutions;
• and
• train a machine-learning model based on a convolutional network, wherein the convolutional neural network comprises the one or more building blocks.

A system comprising:

• one or more processors;
• and a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:
• access a plurality of content objects;
• generate a plurality of voxelized representations for the plurality of content objects, respectively;
• generate, based on one or more sparse convolutions, one or more building blocks, wherein generating each of the one or more building blocks comprises:
• determine, based on the voxelized representation of each of the plurality of content objects, one or more active sites for each of the plurality of content objects;
• and
• apply, to the one or more active sites, the one or more sparse convolutions;
• and
• train a machine-learning model based on a convolutional network, wherein the convolutional neural network comprises the one or more building blocks.