CQUniversity
Browse

Using metamorphic relations to improve accuracy and robustness of deep neural networks

conference contribution
posted on 2025-01-28, 06:29 authored by Kun Qiu, Pak PoonPak Poon, Yu Zhou
When applying metamorphic testing to a deep neural network (DNN), the DNN could have an "acceptable" level of accuracy but it performs poorly against some metamorphic relations (MRs). Such a DNN is considered not robust against these MRs. Improving both accuracy and robustness of a DNN is non-trivial because improving one aspect may adversely affect the other aspect. To alleviate this trade-off problem, we proposed a regularization-based method, in which an optimization function is designed to balance a DNN's accuracy and robustness. Then, we designed a reinforcement-learning-based algorithm to optimize this function. We tested our training method with two datasets (SVHN and CIFAR10), and each dataset with two DNN models. When comparing ours with the other six benchmark methods, we found the DNNs trained with our method have a better balance between accuracy and robustness.

History

Start Page

2

End Page

9

Number of Pages

8

Start Date

2024-09-17

Finish Date

2024-09-17

ISBN-13

9798400711176

Location

Vienna, Austria

Publisher

Association for Computing Machinery

Place of Publication

New York, NY

Peer Reviewed

  • Yes

Open Access

  • No

Era Eligible

  • Yes

Name of Conference

9th ACM International Workshop on Metamorphic Testing

Parent Title

MET '24: Proceedings of the 9th ACM International Workshop on Metamorphic Testing

Usage metrics

    CQUniversity

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC