arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2201.11989cs.LGmath.OC
收藏

使用两时间尺度更新规则训练生成对抗网络时临界批量大小的存在性与估计

Existence and Estimation of Critical Batch Size for Training Generative Adversarial Networks with Two Time-Scale Update Rule

Meiji University(明治大学)

Naoki Sato, Hideaki Iiduka

AI总结:

本文研究常数学习率 TTUR 训练 GAN 时批量大小与训练步数的关系,证明存在使 SFO 复杂度最小的临界批量大小,并用 FID 实验验证理论估计。

摘要:

已有结果表明,使用不同学习率(例如不同的常数学习率或不同的衰减学习率)的两时间尺度更新规则(TTUR)在理论和实践中都有助于训练生成对抗网络(GAN)。此外,在使用 TTUR 训练 GAN 时,不仅学习率很重要,批量大小也很重要,二者都会影响训练所需的步数。本文基于常数学习率,研究批量大小与使用 TTUR 训练 GAN 所需步数之间的关系。我们从理论上证明,对于采用常数学习率的 TTUR,随着批量大小增大,找到判别器和生成器损失函数驻点所需的步数会减少,并且存在一个使随机一阶预言机(SFO)复杂度最小的临界批量大小。随后,我们使用 Fréchet Inception Distance(FID)作为训练性能度量,并给出数值结果,表明达到较低 FID 分数所需的步数会随着批量大小增大而减少;而一旦批量大小超过所测得的临界批量大小,SFO 复杂度就会增加。此外,我们还表明,测得的临界批量大小与根据理论结果估计出的大小接近。

英文摘要:

Previous results have shown that a two time-scale update rule (TTUR) using different learning rates, such as different constant rates or different decaying rates, is useful for training generative adversarial networks (GANs) in theory and in practice. Moreover, not only the learning rate but also the batch size is important for training GANs with TTURs and they both affect the number of steps needed for training. This paper studies the relationship between batch size and the number of steps needed for training GANs with TTURs based on constant learning rates. We theoretically show that, for a TTUR with constant learning rates, the number of steps needed to find stationary points of the loss functions of both the discriminator and generator decreases as the batch size increases and that there exists a critical batch size minimizing the stochastic first-order oracle (SFO) complexity. Then, we use the Fr'echet inception distance (FID) as the performance measure for training and provide numerical results indicating that the number of steps needed to achieve a low FID score decreases as the batch size increases and that the SFO complexity increases once the batch size exceeds the measured critical batch size. Moreover, we show that measured critical batch sizes are close to the sizes estimated from our theoretical results.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑