Graham generalizes sparse convolutional neural networks previously considered in  to 3D data. His approach is twofold:
Graham conducts several experiments meant as proof of concept how these two techniques can be used to speed up 3D convolutional networks, see the paper.
The idea with a tetrahedral grid seems interesting, but concerning the speeded up convolutions, the approach by Engelcke et al.  seems more elegant.
What is your opinion on the summarized work? Or do you know related work that is of interest? Let me know your thoughts in the comments below: