In the carefully controlled small-1D, small-learning-rate, very-wide-network setting, the authors report that optimizing the Fourier mapping through theoretical validation loss also attains the best performance when training actual networks. · CiteArk
In the carefully controlled small-1D, small-learning-rate, very-wide-network setting, the authors report that optimizing the Fourier mapping through theoretical validation loss also attains the best performance when training actual networks.
来源：paper:§5 p. 6, 'Tuning Fourier features in practice'; §A.1 p. 12
