Exponential separations between classical and quantum learners
Casper Gyurik, Vedran Dunjko
TL;DR
The work analyzes when quantum learners provably outperform classical learners for problems with classical data in the PAC framework, clarifying that the source of hardness (identifying a generator versus evaluating a function) crucially shapes the type of quantum advantage. It introduces a refined complexity-theoretic machinery (HeurBPP/samp, HeurP/poly) and multiple hardness assumptions to construct exponential separations, including DLP- and cube-root-based tasks, both with data generation and with fixed hypothesis classes. It further extends to settings without efficient data generation, including data from physical quantum systems, linking learning separations to Hamiltonian learning and order parameters in phases of matter. The paper also connects to prior work on classical shadows and the broader landscape of physically motivated PAC learning, highlighting limitations and conditions under which quantum advantages persist. Overall, the results provide a framework and concrete examples showing that quantum learning can surpass classical methods in a broad range of scenarios, including those motivated by physics.
Abstract
Despite significant effort, the quantum machine learning community has only demonstrated quantum learning advantages for artificial cryptography-inspired datasets when dealing with classical data. In this paper we address the challenge of finding learning problems where quantum learning algorithms can achieve a provable exponential speedup over classical learning algorithms. We reflect on computational learning theory concepts related to this question and discuss how subtle differences in definitions can result in significantly different requirements and tasks for the learner to meet and solve. We examine existing learning problems with provable quantum speedups and find that they largely rely on the classical hardness of evaluating the function that generates the data, rather than identifying it. To address this, we present two new learning separations where the classical difficulty primarily lies in identifying the function generating the data. Furthermore, we explore computational hardness assumptions that can be leveraged to prove quantum speedups in scenarios where data is quantum-generated, which implies likely quantum advantages in a plethora of more natural settings (e.g., in condensed matter and high energy physics). We also discuss the limitations of the classical shadow paradigm in the context of learning separations, and how physically-motivated settings such as characterizing phases of matter and Hamiltonian learning fit in the computational learning framework.