Comment on Every Model Learned by Gradient Descent Is Approximately a Kernel MachineparentComments−QuesnayJr5yThey sketch the argument for SGD, but they don't know if it actually holds (see Remark 5 in the paper).
Comments
They sketch the argument for SGD, but they don't know if it actually holds (see Remark 5 in the paper).