More gradual feature falloff (lower p) improves the theoretical and observed overall training convergence and substantially speeds convergence at mid and high frequencies. · CiteArk
More gradual feature falloff (lower p) improves the theoretical and observed overall training convergence and substantially speeds convergence at mid and high frequencies.
来源：paper:Figures 3(c,d) p. 6 and 12(c,d) p. 15; §5 and §B.1
