Generalized but not Robust? Comparing the Effects of Data Modification Methods on Out-of-Domain Generalization and Adversarial Robustness

被引:0
|
作者
Gokhale, Tejas [1 ]
Mishra, Swaroop [1 ]
Luo, Man [1 ]
Sachdeva, Bhavdeep Singh [1 ]
Baral, Chitta [1 ]
机构
[1] Arizona State Univ, Tempe, AZ 85287 USA
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Data modification, either via additional training datasets, data augmentation, debiasing, and dataset filtering, has been proposed as an effective solution for generalizing to out-of-domain (OOD) inputs, in both natural language processing and computer vision literature. However, the effect of data modification on adversarial robustness remains unclear. In this work, we conduct a comprehensive study of common data modification strategies and evaluate not only their in-domain and OOD performance, but also their adversarial robustness (AR). We also present results on a two-dimensional synthetic dataset to visualize the effect of each method on the training distribution. This work serves as an empirical study towards understanding the relationship between generalizing to unseen domains and defending against adversarial perturbations. Our findings suggest that more data (either via additional datasets or data augmentation) benefits both OOD accuracy and AR. However, data filtering (previously shown to improve OOD accuracy on natural language inference) hurts OOD accuracy on other tasks such as question answering and image classification. We provide insights from our experiments to inform future work in this direction.
引用
收藏
页码:2705 / 2718
页数:14
相关论文
共 9 条
  • [1] Improving Adversarial Robustness via Unlabeled Out-of-Domain Data
    Deng, Zhun
    Zhang, Linjun
    Ghorbani, Amirata
    Zou, James
    24TH INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND STATISTICS (AISTATS), 2021, 130
  • [2] In and Out-of-Domain Text Adversarial Robustness via Label Smoothing
    Yang, Yahan
    Dan, Soham
    Roth, Dan
    Lee, Insup
    61ST CONFERENCE OF THE THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, ACL 2023, VOL 2, 2023, : 657 - 669
  • [3] 3D Adversarial Augmentations for Robust Out-of-Domain Predictions
    Lehner, Alexander
    Gasperini, Stefano
    Marcos-Ramiro, Alvaro
    Schmidt, Michael
    Navab, Nassir
    Busam, Benjamin
    Tombari, Federico
    INTERNATIONAL JOURNAL OF COMPUTER VISION, 2024, 132 (03) : 931 - 963
  • [4] OodGAN: Generative Adversarial Network for Out-of-Domain Data Generation
    Marek, Petr
    Naik, Vishal Ishwar
    Auvray, Vincent
    Goyal, Anuj
    2021 CONFERENCE OF THE NORTH AMERICAN CHAPTER OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, NAACL-HLT 2021, 2021, : 238 - 245
  • [5] ADVERSARIAL GENERATIVE DISTANCE-BASED CLASSIFIER FOR ROBUST OUT-OF-DOMAIN DETECTION
    Zeng, Zhiyuan
    Xu, Hong
    He, Keqing
    Yan, Yuanmeng
    Liu, Sihong
    Liu, Zijun
    Xu, Weiran
    2021 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP 2021), 2021, : 7658 - 7662
  • [6] SSMBA: Self-Supervised Manifold Based Data Augmentation for Improving Out-of-Domain Robustness
    Ng, Nathan
    Cho, Kyunghyun
    Ghassemi, Marzyeh
    PROCEEDINGS OF THE 2020 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (EMNLP), 2020, : 1268 - 1283
  • [7] Creating Robust Children’s ASR System in Zero-Resource Condition Through Out-of-Domain Data Augmentation
    Vinit Kumar
    Avinash Kumar
    S. Shahnawazuddin
    Circuits, Systems, and Signal Processing, 2022, 41 : 2205 - 2220
  • [8] Creating Robust Children's ASR System in Zero-Resource Condition Through Out-of-Domain Data Augmentation
    Kumar, Vinit
    Kumar, Avinash
    Shahnawazuddin, S.
    CIRCUITS SYSTEMS AND SIGNAL PROCESSING, 2022, 41 (04) : 2205 - 2220
  • [9] Selecting Augmentation Methods for Domain Generalization and Out-of-Distribution Detection Using Unlabeled Data
    Kucuktas, Ulku Tuncer
    Uysal, Fatih
    Hardalac, Firat
    32ND IEEE SIGNAL PROCESSING AND COMMUNICATIONS APPLICATIONS CONFERENCE, SIU 2024, 2024,