Dynamic Stochastic Ensemble with Adversarial Robust Lottery Ticket Subnetworks Qi Peng1Wenlin Liu2Ruoxi Qin1Libin Hou1Bin Yan1Linyuan Wang1

2025-05-03 1 0 815.43KB 6 页 10玖币

侵权投诉

Dynamic Stochastic Ensemble with Adversarial

Robust Lottery Ticket Subnetworks

Qi Peng,1Wenlin Liu,2Ruoxi Qin,1Libin Hou,1Bin Yan,1Linyuan Wang1∗

1PLA Strategy Support Force Information Engineering University

2University of Science and Technology of China (USTC)

pqmailkwiki@163.com, wanglinyuanwly@163.com

Abstract

Adversarial attacks are considered the intrinsic vulnerability of CNNs. Defense

strategies designed for attacks have been stuck in the adversarial attack-defense

arms race, reﬂecting the imbalance between attack and defense. Dynamic Defense

Framework (DDF) recently changed the passive safety status quo based on the

stochastic ensemble model. The diversity of subnetworks, an essential concern in

the DDF, can be effectively evaluated by the adversarial transferability between

different networks. Inspired by the poor adversarial transferability between sub-

networks of scratch tickets with various remaining ratios, we propose a method

to realize the dynamic stochastic ensemble defense strategy. We discover the

adversarial transferable diversity between robust lottery ticket subnetworks drawn

from different basic structures and sparsity. The experimental results suggest that

our method achieves better robust and clean recognition accuracy by adversarial

transferable diversity, which would decrease the reliability of attacks.

1 Introduction

Deep neural networks (DNNs) currently deﬁne state-of-the-art performance in standard image

classiﬁcation tasks. However, Szegedy et al. proposed that the advanced classiﬁers may be fooled

by an imperceptible perturbation called adversarial samples[

]. It raises concern about the intrinsic

vulnerability of DNNs[2, 3].

The defenders try hard to gain the initiative in the adversarial arms race to resist the rapid development

of adversarial attacks. Researchers propose many empirical and certiﬁed defense methods to obtain

robust networks. The certiﬁed defense methods are supported by rigorous theoretical security

guarantees, which could steadily expand the robustness radius. However, transferring it to large

datasets is not accessible due to the high computational cost[

]. And adversarial training is currently

the most ﬂexible and effective empirical defense method by enhancing the training set with adversarial

samples generated dynamically[

]. Nevertheless, previous research has testiﬁed the widespread

transferability of adversarial examples[

]. And there is a demand for implicit transferability in

further research because adversarial training depends on speciﬁc attack algorithms for augmented

data[

], which makes the defender hard to and appear passive in the arms race. On the contrary,

a rational ensemble strategy is an effective defense method in practice[

]. A recent study

presents the dynamic defense framework (DDF) based on the stochastic ensemble[

]. The DDF

would change the ensemble states based on variable model attributes of the architecture and smooth

parameters. It expects heterogeneous candidate models to ensure diverse ensemble statuses.

We propose the dynamic stochastic ensemble with adversarial robust lottery ticket subnetworks.

Based on the Lottery Hypothesis[

], Fu et al. discover the subnetworks with inborn robustness. It

∗Corresponding author.

Preprint. Under review.

arXiv:2210.02618v1 [cs.CV] 6 Oct 2022

matches or surpasses the adversarially trained networks with the same structure without any model

training.[

]. Inspired by Fu et al., our method obtained subnetworks with different network structures

and remaining ratios to promote the adversarial transferable diversity for the DDF. By weakening the

transferability between ensemble states, we improve the initiative of the DDF against the adversary.

2 Method

In the framework of dynamic defense, we represent the dynamic stochastic ensemble with adversarial

robust lottery ticket subnetworks. [

] proved the poor adversarial transferability between the scratch

tickets under a single structure. Drawing inspiration from prior works, we further explore the

adversarial transferable diversity from the different fundamental structures and remaining ratios.

2.1 The Dynamic Defense Framework and Adversarial Transferable Diversity

The DDF is a randomized defense strategy to protect ensemble gradient information, and the essential

requirements for it are randomness and diversity to promote the ensemble’s adversarial robustness[

It presents a model ensemble defense method with randomized network parameter distribution

specialty, which causes an unknowable act of the defender. The output of dynamic stochastic

ensemble model fens containing Inumber of models is deﬁned as follows:

fens(x, θ) =

i=1

f(x, θ)(1)

The randomness is achieved by transferring the ensemble states with ensemble variables

. The

DDF demands the construction of diversiﬁed ensemble statuses with a heterogeneous model library.

Relevant studies highlight that diverse network structure plays a crucial role in ensemble defense[

In our solution, we evaluated the heterogeneousness and diversity between ensemble subnetworks by

the poor adversarial transferability of the attacks.

2.2 Adversarial Robust Lottery Subnetwork

For purpose of testifying the multi-sparsity adversarial robust lottery subnetworks can achieve better

adversarial transferable diversity under different network structures. We picked four representative

network structures, ResNet18, ResNet34, WideResNet32, and WideResNet38[

], as the basic

architecture of our experiments and gained the sparse lottery ticket from original dense networks.

Following [

], we applied adversarial training to gain robustness of our subnetworks during pruning.

It can be expressed as a min-max problem as Eq.2.

arg min

λX

max

kδk≤εl(f(ˆ

λω, xi+δ), yi)s.t. kˆ

λk0≤k(2)

Where lpresents the loss function, fis a randomly initialized network with random weights, and

is the perturbation with maximum value

. In order to satisfy the sparsity of the subnetworks, we

set a learnable weight

and a binary weight

λ∈ {0,1}

that correspond to its dimensions[

is meant to activate a small number of primary weights

. With the primary network parameters

weighted by λ∈(0,1),fcan be effectively trained by small perturbations added to the input xi.

2.3 Dynamical Ensemble for The Lottery Subnetworks

Through our method, we obtained fourty subnetworks with different basic structures and sparsity.

Based on the robust lottery subnetwork library, we deﬁne the randomized ensemble attribute parameter

θ=θ(α, n, s), which determines the ensemble states. It can be achieved in the following steps:

(A)Construct a robust lottery subnetworks library with adversarial transferable diversity, including

forty sparse subnetworks. Each of ResNet18/ResNet34/WideResNet32/WideResNet38 owns ten.

(B)Set the range for

and s. We brought four basic structures into the selection rather than

the entire library, increasing the possibility of including more structures. It realized by

α=

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

DynamicStochasticEnsemblewithAdversarialRobustLotteryTicketSubnetworksQiPeng,1WenlinLiu,2RuoxiQin,1LibinHou,1BinYan,1LinyuanWang11PLAStrategySupportForceInformationEngineeringUniversity2UniversityofScienceandTechnologyofChina(USTC)pqmailkwiki@163.com,wanglinyuanwly@163.comAbstractAdversarialattacks...

展开>> 收起<<

Dynamic Stochastic Ensemble with Adversarial Robust Lottery Ticket Subnetworks Qi Peng1Wenlin Liu2Ruoxi Qin1Libin Hou1Bin Yan1Linyuan Wang1.pdf

共6页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Dynamic Stochastic Ensemble with Adversarial Robust Lottery Ticket Subnetworks Qi Peng1Wenlin Liu2Ruoxi Qin1Libin Hou1Bin Yan1Linyuan Wang1

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: