Efficient Function Approximation Under Heteroskedastic Noise

Yuji Nakatsukasa, Yifu Zhang · 2025

Read Efficient Function Approximation Under Heteroskedastic Noise by Yuji Nakatsukasa, Yifu Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-08683.txt.

Description

Yuji Nakatsukasa and Yifu Zhang present a 2025 numerical analysis paper on fitting a smooth curve to data whose noise level changes from place to place. When measurement error varies along the interval, standard fitting wastes samples where noise is low and drowns where it is high. No fast algorithm fully handled this uneven case.

Their method, HeteroChebtrunc, adapts a Chebyshev-polynomial fitting routine to weight its effort by local noise. Probability arguments show it beats its predecessor's worst-case error while running in near-linear time. Along the way the authors prove a sharp bound on how well sample variance itself can be estimated, a result they expect to travel. Numerical tests confirm the tighter fit.

Catalog record

CREATORYuji Nakatsukasa, Yifu Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-08683

Browse the full library · About the Tower of Babel