'Are weights of a trained network based on ImageNet a derived work, a cesspool of million of copyright claims?'
I'd argue not. The success of deep learning in vision seems to be in acquiring allocentric representations of objects. Copyright protects the expression and not the concept. Parameter weights describe the concept, not particular instantiations of it.
This is what I am hoping too - but derived works are a sticky subject. Imagine a scenario where someone gets of one of these "data vaults" from a large company, then trains a network and throws away the training data. You are still holding the "essence" of their datastore, even without the actual data.
I guess we won't really know until someone goes to court over it.
I'd argue not. The success of deep learning in vision seems to be in acquiring allocentric representations of objects. Copyright protects the expression and not the concept. Parameter weights describe the concept, not particular instantiations of it.