Comment on Every Model Learned by Gradient Descent Is Approximately a Kernel MachineparentComments−riku_iki5yIt has been training that approximator.Also it is representation of approximator itself and data compression. For cubic splines to approximate some NN you likely would need enormous number of intervals covering input space.−VHRanger5yAlso, cubic splines don't extrapolate like the other mentioned models
Comments
Also it is representation of approximator itself and data compression. For cubic splines to approximate some NN you likely would need enormous number of intervals covering input space.
Also, cubic splines don't extrapolate like the other mentioned models