Comparing the Effects of Boltzmann Machines as Associative Memory in Generative Adversarial Networks between Classical and Quantum Samplings
Comparing the Effects of Boltzmann Machines as Associative Memory in Generative Adversarial Networks between Classical and Quantum Samplings
复制标题
DOI：
10.7566/jpsj.91.074008
复制
发表时间：
2022-07-15
影响因子：
1.7
通讯作者：
Tanaka, Kazuyuki
中科院分区：
文献类型：
--
作者：
Urushibata, Mitsuru;Ohzeki, Masayuki;Tanaka, Kazuyuki
我们研究了量子效应对机器学习（ML）模型的影响，例如生成对抗网络（GAN），这是一个很有前途的深度学习框架。在一般的GAN框架中，生成器将均匀噪声映射到假图像。在这项研究中，我们利用了联想对抗网络（AAN），它由标准GAN和联想记忆组成。此外，我们设置了一个玻尔兹曼机（BM），这是一个无向图形模型，学习低维特征提取的一个矩阵，作为内存。由于BM的对数似然梯度难以计算，因此需要用BM的样本均值来近似计算。为了计算样本平均值，通常使用马尔可夫链蒙特卡罗（MCMC）方法。在之前的研究中，这是使用量子退火设备进行的，并将“量子”AAN的性能与标准GAN的性能进行了比较。然而，它比标准GAN更好的性能还没有得到很好的理解。在这项研究中，我们介绍了两种方法来提取样本：经典采样通过MCMC和量子采样通过量子蒙特卡罗（QMC）模拟，这是量子模拟在经典计算机上。然后，我们比较这些方法，以调查量子采样是否是有利的。具体来说，计算的反射损耗，发电机损耗，起始分数，和Frechet起始距离，我们讨论了AAN的可能性。我们表明，经过MCMC和QMC训练的AAN在训练过程中比标准GAN更稳定，并且产生更多不同的图像。然而，结果表明，与MCMC相比，QMC模拟的采样没有差异。
We investigate the quantum effect on machine learning (ML) models exemplified by the Generative Adversarial Network (GAN), which is a promising deep learning framework. In the general GAN framework, the generator maps uniform noise to a fake image. In this study, we utilize the Associative Adversarial Network (AAN), which consists of a standard GAN and an associative memory. Moreover, we set a Boltzmann Machine (BM), which is an undirected graphical model that learns low-dimensional features extracted from a discriminator, as the memory. Owing to the difficulty of calculating the BM's log-likelihood gradient, it is necessary to approximate it by using the sample mean obtained from the BM, which has tentative parameters. To calculate the sample mean, a Markov Chain Monte Carlo (MCMC) method is often used. In a previous study, this was performed using a quantum annealing device, and the performance of the "Quantum" AAN was compared with that of the standard GAN. However, its better performance than the standard GAN is not well understood. In this study, we introduce two methods to draw samples: classical sampling via MCMC and quantum sampling via quantum Monte Carlo (QMC) simulation, which is quantum simulation on the classical computer. Then, we compare these methods to investigate whether quantum sampling is advantageous. Specifically, calculating the discriminator loss, generator loss, inception score, and Frechet inception distance, we discuss the possibility of AAN. We show that the AANs trained by both MCMC and QMC, are more stable during training and produce more varied images than the standard GANs. However, the results indicate no difference in sampling by QMC simulation compared with that by MCMC.