[论文解读] Generalization Metrics for Practical Quantum Advantage in Generative Models

Kaitlin Gili, Marta Mauri+1 more|arXiv (Cornell University)|Jan 21, 2022

Quantum Computing Algorithms and Architecture被引用 6

一句话总结

本文提出了一种统一的、基于样本的框架，通过在离散的、受限的优化数据集上评估泛化性能，来衡量生成建模中的实际量子优势。结果表明，基于量子启发张量网络的模型（TNBMs）生成的唯一且有效的未见样本数量是 GAN 的 68 倍，高质量样本的性能比 GAN 高 61.2 倍，为评估真实世界生成任务中量子优势提供了一项稳健且可比较的指标。

ABSTRACT

As the quantum computing community gravitates towards understanding the practical benefits of quantum computers, having a clear definition and evaluation scheme for assessing practical quantum advantage in the context of specific applications is paramount. Generative modeling, for example, is a widely accepted natural use case for quantum computers, and yet has lacked a concrete approach for quantifying success of quantum models over classical ones. In this work, we construct a simple and unambiguous approach to probe practical quantum advantage for generative modeling by measuring the algorithm's generalization performance. Using the sample-based approach proposed here, any generative model, from state-of-the-art classical generative models such as GANs to quantum models such as Quantum Circuit Born Machines, can be evaluated on the same ground on a concrete well-defined framework. In contrast to other sample-based metrics for probing practical generalization, we leverage constrained optimization problems (e.g., cardinality-constrained problems) and use these discrete datasets to define specific metrics capable of unambiguously measuring the quality of the samples and the model's generalization capabilities for generating data beyond the training set but still within the valid solution space. Additionally, our metrics can diagnose trainability issues such as mode collapse and overfitting, as we illustrate when comparing GANs to quantum-inspired models built out of tensor networks. Our simulation results show that our quantum-inspired models have up to a $68 imes$ enhancement in generating unseen unique and valid samples compared to GANs, and a ratio of 61:2 for generating samples with better quality than those observed in the training set. We foresee these metrics as valuable tools for rigorously defining practical quantum advantage in the domain of generative modeling.

研究动机与目标

• 建立一个清晰、统一的框架，用于衡量生成建模中的实际量子优势，区别于渐近量子加速。
• 定义并量化泛化性能，使其适用于经典和量子生成模型。
• 通过基于样本的度量诊断常见的训练问题，如模式崩溃和过拟合。
• 在相同评估基准上实现对经典模型（如 GAN）和量子启发模型（如 TNBMs）的公平、直接比较。
• 为未来在真实世界应用场景中评估量子和经典生成模型提供可扩展、实用的基准。

提出的方法

• 该框架使用离散的、受限的优化问题（如基数约束问题）作为数据集，以定义一个明确且无歧义的解空间。
• 通过评估模型生成的样本在解空间内且未在训练集中出现的能力来衡量泛化性能。
• 关键度量包括在训练集之外生成的唯一有效样本数量，以及优于训练集中任意样本的高质量样本占总生成样本的比例。
• 该方法利用基于样本的评估来检测训练陷阱（如模式崩溃和过拟合），通过分析生成输出的多样性与有效性。
• 该方法应用于在相同离散数据集上对经典 GAN 和量子启发张量网络模型（TNBMs）进行比较，采用模拟评估。
• 该框架设计为可扩展至未来的混合架构和全量子生成架构，包括量子线路玻恩机（Quantum Circuit Born Machines）。

Figure 1: A visual representation of generalization-related concepts. The figure shows the different behaviours a model can exhibit when generating data, using a 3x3 Bars and Stripes dataset as an example. The top two rows display a set of samples $x$ distributed according to the data distribution $

实验结果

研究问题

• RQ1 如何以一种独立于渐近尺度、适用于现实世界问题的方式，定义并衡量生成建模中的实际量子优势？
• RQ2 在多大程度上，将泛化性能（即生成多样、有效且高质量的、超出训练集的样本）作为实际量子优势的可靠度量？
• RQ3 是否可以建立一个统一的评估框架，公平地比较经典生成模型（如 GAN）和量子启发模型（如 TNBMs）在相同离散数据集上的表现？
• RQ4 常见的训练问题（如模式崩溃和过拟合）在不同生成模型的泛化性能中如何表现？
• RQ5 与最先进的经典模型相比，量子启发模型在泛化性能上可实现哪些可量化的改进？

主要发现

• 在相同离散数据集上，量子启发张量网络模型（TNBMs）生成的唯一且有效未见样本数量是 GAN 的 68 倍。
• TNBMs 生成的高质量样本（优于训练集中任意样本）与 GAN 生成的高质量样本之比为 61.2，表明其质量优势显著。
• 该框架通过分析生成样本的多样性和有效性，成功检测到 GAN 中的模式崩溃和过拟合现象。
• 本研究证明，泛化性能是生成建模中实际量子优势的强而可测量的代理指标。
• 所提出的度量方法稳健、可扩展，适用于经典和量子生成模型，实现了公平且直接的比较。
• 该框架可扩展至未来的量子硬件和混合架构，为在真实量子设备上评估泛化性能提供了路径。

Figure 2: A visual representation of all possible spaces where a generated query might be located. Each query is represented by a color-coded dot, where the color-code is the same as in Figure 1 (Data-Copying: blue, Pre-Generalization: yellow, Validity-Based Generalization: purple) and the color-sha