a lot of the models i have worked with are embedding models

we have a model that works in terms of a single and also hierarchical and or some other structure

embedding models work because they build

however if you apply some hierarchical embedding you want to add some regularization to the layers

otherwise the model has no incentive to put anything in the higher levels

irt one dimensional ca higher dimensional nn higher dimensional but restricted

the models are quite simple, but they power a whole platform

simple model with lots of data beats a complex model with no data