Saved in:
Bibliographic Details
Main Author: Zanger, Daniel Z.
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.15576
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917644388532224
author Zanger, Daniel Z.
author_facet Zanger, Daniel Z.
contents Using quantum algorithms, we obtain, for accuracy $ε>0$ and confidence $1-δ,0<δ<1,$ a new sample complexity upper bound of $O((\mbox{log}(\frac{1}δ))/ε)$ as $ε,δ\rightarrow 0$ for a general agnostic learning model, provided the hypothesis class is of finite cardinality. This greatly improves upon a corresponding sample complexity of asymptotic order $Θ((\mbox{log}(\frac{1}δ))/ε^{2})$ known in the literature to be attainable by means of classical (non-quantum) algorithms for an agnostic learning problem also with hypothesis set of finite cardinality (see, for example, Arunachalam and de Wolf (2018) and the classical statistical learning theory references cited there). Thus, for general agnostic learning, the quantum speedup in the rate of learning that we achieve with respect to these results is quadratic in $ε^{-1}$.
format Preprint
id arxiv_https___arxiv_org_abs_2310_15576
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle A Quadratic Sample Complexity Reduction for Agnostic Learning via Quantum Algorithms
Zanger, Daniel Z.
Quantum Physics
Using quantum algorithms, we obtain, for accuracy $ε>0$ and confidence $1-δ,0<δ<1,$ a new sample complexity upper bound of $O((\mbox{log}(\frac{1}δ))/ε)$ as $ε,δ\rightarrow 0$ for a general agnostic learning model, provided the hypothesis class is of finite cardinality. This greatly improves upon a corresponding sample complexity of asymptotic order $Θ((\mbox{log}(\frac{1}δ))/ε^{2})$ known in the literature to be attainable by means of classical (non-quantum) algorithms for an agnostic learning problem also with hypothesis set of finite cardinality (see, for example, Arunachalam and de Wolf (2018) and the classical statistical learning theory references cited there). Thus, for general agnostic learning, the quantum speedup in the rate of learning that we achieve with respect to these results is quadratic in $ε^{-1}$.
title A Quadratic Sample Complexity Reduction for Agnostic Learning via Quantum Algorithms
topic Quantum Physics
url https://arxiv.org/abs/2310.15576