In this paper it is shown that $C_\beta$-smooth functions can be approximated by deep neural networks with ReLU activation function and with parameters $\{0,\pm \frac{1}{2}, \pm 1, 2\}$. The $l_0$ and $l_1$ parameter norms of considered networks are thus equivalent. The depth, width and the number of active parameters of the constructed networks have, up to a logarithmic factor, the same dependence on the approximation error as the networks with parameters in $[-1,1]$. In particular, this means that the nonparametric regression estimation with the constructed networks attains the same convergence rate as with sparse networks with parameters in $[-1,1]$.
翻译:本文显示,具有RELU激活功能的深神经网络和参数为 $0,\pm\frac{1 ⁇ 2}、\ pm 1, 2 ⁇ 2}的参数可以近似于 $C ⁇ beta$-smoth 函数。因此,考虑的网络的参数规范是等同的。所建网络的深度、宽度和有效参数数量,在对数系数上,对近似误差的依赖程度与以 $[1,1,1]为参数的网络相同。这特别意味着,对已建网络的非参数回归估计达到与以 $[1,1,1]为参数的稀少网络相同的趋同率。