4 ms·
What do you mean with more insights? If they are not represented as some data, how could they ever influence any computational model?
by psychoslave 3y ago
What do you mean with more insights? If they are not represented as some data, how could they ever influence any computational model?
- davegirdywww 3y agoNobody said anything about insights.
- niemandhier 3y agoBy building the model in such a way that assumptions about the data are in the model structure. ConvNets hardcode translation equivariance, more general convolutions can hardcode more general equivariant structures. The whole field of geometric deep learning is about constructing nets based on insights about the structure of the data manifold, and many successful nets turn out to do this.