Finding Deviated Behaviors of the Compressed DNN Models for Image Classifications

Author:

Tian Yongqiang1ORCID,Zhang Wuqi2ORCID,Wen Ming3ORCID,Cheung Shing-Chi2ORCID,Sun Chengnian4ORCID,Ma Shiqing5ORCID,Jiang Yu6ORCID

Affiliation:

1. University of Waterloo, Canada and The Hong Kong University of Science and Technology, China

2. The Hong Kong University of Science and Technology, China

3. Huazhong University of Science and Technology, China

4. University of Waterloo, Canada

5. Rutgers University, USA

6. Tsinghua University, China

Abstract

Model compression can significantly reduce the sizes of deep neural network (DNN) models and thus facilitate the dissemination of sophisticated, sizable DNN models, especially for deployment on mobile or embedded devices. However, the prediction results of compressed models may deviate from those of their original models. To help developers thoroughly understand the impact of model compression, it is essential to test these models to find those deviated behaviors before dissemination. However, this is a non-trivial task, because the architectures and gradients of compressed models are usually not available. To this end, we propose Dflare , a novel, search-based, black-box testing technique to automatically find triggering inputs that result in deviated behaviors in image classification tasks. Dflare iteratively applies a series of mutation operations to a given seed image until a triggering input is found. For better efficacy and efficiency, Dflare models the search problem as Markov Chains and leverages the Metropolis-Hasting algorithm to guide the selection of mutation operators in each iteration. Further, Dflare utilizes a novel fitness function to prioritize the mutated inputs that either cause large differences between two models’ outputs or trigger previously unobserved models’ probability vectors. We evaluated Dflare on 21 compressed models for image classification tasks with three datasets. The results show that Dflare not only constantly outperforms the baseline in terms of efficacy but also significantly improves the efficiency: Dflare is 17.84×–446.06× as fast as the baseline in terms of time; the number of queries required by Dflare to find one triggering input is only 0.186–1.937% of those issued by the baseline. We also demonstrated that the triggering inputs found by Dflare can be used to repair up to 48.48% deviated behaviors in image classification tasks and further decrease the effectiveness of Dflare on the repaired models.

Funder

National Natural Science Foundation of China

Hong Kong RGC/GRF

Hong Kong RGC/RIF

Hong Kong ITF

Hong Kong PhD Fellowship Scheme, HKUST RedBird Academic Excellence Award, and the MSRA Collaborative Research Grant

Cisco Research Gift, Natural Sciences and Engineering Research Council of Canada (NSERC) through the Discovery Grant, and CFI-JELF Project

Publisher

Association for Computing Machinery (ACM)

Subject

Software

Reference82 articles.

1. Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey

2. Pearson Correlation Coefficient

3. Arjun Nitin Bhagoji, Warren He, Bo Li, and Dawn Song. 2018. Practical black-box attacks on deep neural networks using efficient query mechanisms. In ECCV’18, Vol. 11216. Springer, 158–174.

4. Image Noise Models

5. Model compression

Cited by 1 articles. 订阅此论文施引文献 订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献

1. CoopHance: Cooperative Enhancement for Robustness of Deep Learning Systems;Proceedings of the 32nd ACM SIGSOFT International Symposium on Software Testing and Analysis;2023-07-12

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3