arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2406.04070cs.LGcs.AI
收藏

Batch-in-Batch:一种面向初始扰动与样本选择的新的对抗训练框架

Batch-in-Batch: a new adversarial training framework for initial perturbation and sample selection

School of Mathematics and Computer Science, Jianghan University(江汉大学数学与计算机科学学院);School of Cyber Science and Engineering, Wuhan University(武汉大学网络空间安全学院)

Yinting Wu, Pai Peng, Bo Cai, Le Li, .

AI总结:

本文提出 Batch-in-Batch 对抗训练框架,通过联合生成多组初始扰动并结合样本选择策略提升模型鲁棒性,在多个数据集上显著提高对抗准确率且计算成本较低。

摘要:

对抗训练方法通常从简单的均匀分布中为对抗样本生成相互独立的初始扰动,并在不进行选择的情况下获得用于分类器的训练批次。在这项工作中,我们提出了一个简单而有效的训练框架,称为 Batch-in-Batch(BB),以增强模型的鲁棒性。它具体涉及一种初始值的联合构造方式,该方式能够从原始批次集合中同时生成 $m$ 组扰动,从而为对抗样本提供更多的多样性;同时还包括多种样本选择策略,使训练后的模型具有更平滑的损失并避免过度自信的输出。通过在三个基准数据集(CIFAR-10、SVHN、CIFAR-100)上使用两个网络(PreActResNet18 和 WideResNet28-10)进行的大量实验,这些网络同时用于单步(Noise-Fast Gradient Sign Method,N-FGSM)和多步(Projected Gradient Descent,PGD-10)对抗训练,我们表明在 BB 框架内训练的模型在各种对抗设置下始终具有更高的对抗准确率,尤其是在攻击半径为 8/255 的 SVHN 数据集上,与 N-FGSM 基线模型相比实现了超过 13% 的提升。此外,对所提出的初始扰动方法和样本选择策略的效率进行的实验分析验证了我们的见解。最后,我们表明,即使在 $m$ 值相对较大的情况下,我们的框架在计算资源方面也具有成本效益。

英文摘要:

Adversarial training methods commonly generate independent initial perturbation for adversarial samples from a simple uniform distribution, and obtain the training batch for the classifier without selection. In this work, we propose a simple yet effective training framework called Batch-in-Batch (BB) to enhance models robustness. It involves specifically a joint construction of initial values that could simultaneously generates $m$ sets of perturbations from the original batch set to provide more diversity for adversarial samples; and also includes various sample selection strategies that enable the trained models to have smoother losses and avoid overconfident outputs. Through extensive experiments on three benchmark datasets (CIFAR-10, SVHN, CIFAR-100) with two networks (PreActResNet18 and WideResNet28-10) that are used in both the single-step (Noise-Fast Gradient Sign Method, N-FGSM) and multi-step (Projected Gradient Descent, PGD-10) adversarial training, we show that models trained within the BB framework consistently have higher adversarial accuracy across various adversarial settings, notably achieving over a 13% improvement on the SVHN dataset with an attack radius of 8/255 compared to the N-FGSM baseline model. Furthermore, experimental analysis of the efficiency of both the proposed initial perturbation method and sample selection strategies validates our insights. Finally, we show that our framework is cost-effective in terms of computational resources, even with a relatively large value of $m$.

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。

↑