Current browse context:
cs
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Learning low-degree functions from a logarithmic number of random queries
(Submitted on 21 Sep 2021 (v1), last revised 9 Mar 2022 (this version, v3))
Abstract: We prove that every bounded function $f:\{-1,1\}^n\to[-1,1]$ of degree at most $d$ can be learned with $L_2$-accuracy $\varepsilon$ and confidence $1-\delta$ from $\log(\tfrac{n}{\delta})\,\varepsilon^{-d-1} C^{d^{3/2}\sqrt{\log d}}$ random queries, where $C>1$ is a universal finite constant.
Submission history
From: Alexandros Eskenazis [view email][v1] Tue, 21 Sep 2021 13:19:04 GMT (9kb,D)
[v2] Wed, 3 Nov 2021 09:20:08 GMT (10kb,D)
[v3] Wed, 9 Mar 2022 12:31:50 GMT (10kb,D)
Link back to: arXiv, form interface, contact.