Approximation of functions with one-bit neural networks

TL;DR

This paper investigates the approximation power of one-bit neural networks, demonstrating they can approximate smooth functions with quantized parameters efficiently, and introduces new theoretical results and implementations for such networks.

Contribution

The paper provides the first approximation results for one-bit neural networks, including explicit bounds and novel methods for implementing Bernstein polynomials with quantized parameters.

Findings

One-bit neural networks can approximate smooth functions with controlled error.

Explicit bounds on the number of parameters needed for approximation.

New methods for implementing Bernstein polynomials with quantized neural networks.

Abstract

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.