1: \begin{abstract}

2: Approximating a univariate function on the interval $[-1,1]$ with a polynomial is among the most classical problems in numerical analysis. 

3: When the function evaluations come with noise, a least-squares fit is known to reduce the effect of noise as more samples are taken. 

4: The generic algorithm for the least-squares problem requires $O(Nn^2)$ operations, where $N+1$ is the number of sample points and $n$ is the degree of the polynomial approximant.

5: This algorithm is unstable when $n$ is large, for example $n\gg \sqrt{N}$ for equispaced sample points. In this study, we blend numerical analysis and statistics to introduce a stable and fast $O(N\log N)$ algorithm called \mainalg\ based on the Chebyshev interpolation.

6: It has the same error reduction effect as least-squares and the convergence is spectral until the error reaches $O(\sigma \sqrt{{n}/{N}})$, where $\sigma$ is the noise level, after which the error continues to decrease at the Monte-Carlo $O(1/\sqrt{N})$ rate. 

7: To determine the polynomial degree, \mainalg\ employs a statistical criterion, namely Mallows' $C_p$. 

8: We analyze \mainalg\ in terms of the variance and concentration in the infinity norm to the underlying noiseless function. These results show that with high probability the infinity-norm error is bounded by a small constant times $\sigma \sqrt{{n}/{N}}$, when the noise {is} independent and follows a subgaussian or subexponential distribution.

9: We illustrate the performance of \mainalg\ with numerical experiments. 

10: \end{abstract}

11: