The celebrated universal approximation theorems for neural networks roughly state that any reasonable function can be arbitrarily well-approximated by a network whose parameters are appropriately chosen real numbers. This paper examines the approximation capabilities of one-bit neural networks -- those whose nonzero parameters are $\pm a$ for some fixed $a\not=0$. One of our main theorems shows that for any $f\in C^s([0,1]^d)$ with $\|f\|_\infty<1$ and error $\varepsilon$, there is a $f_{NN}$ such that $|f(\boldsymbol{x})-f_{NN}(\boldsymbol{x})|\leq \varepsilon$ for all $\boldsymbol{x}$ away from the boundary of $[0,1]^d$, and $f_{NN}$ is either implementable by a $\{\pm 1\}$ quadratic network with $O(\varepsilon^{-2d/s})$ parameters or a $\{\pm \frac 1 2 \}$ ReLU network with $O(\varepsilon^{-2d/s}\log (1/\varepsilon))$ parameters, as $\varepsilon\to0$. We establish new approximation results for iterated multivariate Bernstein operators, error estimates for noise-shaping quantization on the Bernstein basis, and novel implementation of the Bernstein polynomials by one-bit quadratic and ReLU neural networks.
翻译:神经网络的普遍逼近定理大致表明,可以通过适当选择参数为实数的网络来任意好地逼近任何合理函数。本文研究了单比特神经网络的逼近能力,即那些非零参数为 $\pm a$ (其中 $a \neq 0$)的神经网络。我们的主要定理之一表明,对于任意 $f\in C^s([0,1]^d)$,其中 $\|f\|_\infty<1$,并且误差为 $\varepsilon$,存在一个 $f_{NN}$,使得所有 $\boldsymbol{x}$ 远离 $[0,1]^d$ 的边界上都有 $|f(\boldsymbol{x})-f_{NN}(\boldsymbol{x})|≤\varepsilon$,且 $f_{NN}$ 可以是可实现的 $\{\pm1\}$ 二次网络或 $\{\pm\frac{1}{2}\}$ ReLU 网络,其中前者的参数数量为 $O(\varepsilon^{-2d/s})$,后者的参数数量为 $O(\varepsilon^{-2d/s}\log(1/\varepsilon))$,当 $\varepsilon \to 0$ 时。我们建立了新的多元 Bernstein 迭代算子的逼近结果,Bernstein 基础上的噪声整形量化误差估计以及 Bernstein 多项式在单比特二次和 ReLU 神经网络中的新型实现。