Sum and product kernels assume the gram matrix is dense.
Propose to exploit the additive gram structure for sum kernels, the Hadamard product structure for product kernels.
The first is easy to implement. The latter might require ensuring the __mul__ method is defined for all linear operators over in JaxLinOp (if not implemented already).
Sum and product kernels assume the gram matrix is dense.
Propose to exploit the additive gram structure for sum kernels, the Hadamard product structure for product kernels.
The first is easy to implement. The latter might require ensuring the
__mul__
method is defined for all linear operators over inJaxLinOp
(if not implemented already).