1. NCCU Academic Hub
  2. Publications
  3. College of Commerce
  4. Periodical Articles
  5. Item

Publications-Periodical Articles

Article View/Open

Publication Export

Google ScholarTM

NCCU Library

Citation Infomation

Related Publications in TAIR

題名 Controllable Robustness Training
作者 郁方
Hu, Yu-Chi;Yu, Fang
貢獻者 資管系
關鍵詞 abstract interpretation; adversarial training; logic rule; robustness; neural network
日期 2025-05
上傳時間 27-Aug-2026 10:32:18 (UTC+8)
摘要 Neural network techniques allow for the developing of complex systems that are dif-ficult for humans to implement. However, they are known to be vulnerable to adversarial examples, where crafted perturbations can change prediction decisions. Training these net-works using adversarial examples or abstraction interpretation can improve robustness but may reduce precision and training performance on the original task prediction. To balance the trade-off between accuracy and robustness, we propose controllable robustness training, where we integrate controllable neural network model with rule representations for robust-ness training process. The loss on adversarial training can then be considered as a loss on the rule, thus separating the robustness training from the original task process. Rule strength can be adjusted at a testing time on its loss ratio without model retraining, which balances precision and robustness in how the model learns rules and constraints. We demonstrate that controlling the contribution of robustness training achieves a better bal-ance of good performance in both the accuracy and robustness of neural networks against various adversarial attacks and perturbations.
關聯 Journal of Information Science and Engineering, Vol.41, No.3, pp.743-768
資料類型 article
DOI https://doi.org/10.6688/JISE.202505_41(3).0014
dc.contributor 資管系
dc.creator (作者) 郁方
dc.creator (作者) Hu, Yu-Chi;Yu, Fang
dc.date (日期) 2025-05
dc.date.accessioned 27-Aug-2026 10:32:18 (UTC+8)-
dc.date.available 27-Aug-2026 10:32:18 (UTC+8)-
dc.date.issued (上傳時間) 27-Aug-2026 10:32:18 (UTC+8)-
dc.identifier.uri (URI) https://ah.lib.nccu.edu.tw/item?item_id=184621-
dc.description.abstract (摘要) Neural network techniques allow for the developing of complex systems that are dif-ficult for humans to implement. However, they are known to be vulnerable to adversarial examples, where crafted perturbations can change prediction decisions. Training these net-works using adversarial examples or abstraction interpretation can improve robustness but may reduce precision and training performance on the original task prediction. To balance the trade-off between accuracy and robustness, we propose controllable robustness training, where we integrate controllable neural network model with rule representations for robust-ness training process. The loss on adversarial training can then be considered as a loss on the rule, thus separating the robustness training from the original task process. Rule strength can be adjusted at a testing time on its loss ratio without model retraining, which balances precision and robustness in how the model learns rules and constraints. We demonstrate that controlling the contribution of robustness training achieves a better bal-ance of good performance in both the accuracy and robustness of neural networks against various adversarial attacks and perturbations.
dc.format.extent 145 bytes-
dc.format.mimetype text/html-
dc.relation (關聯) Journal of Information Science and Engineering, Vol.41, No.3, pp.743-768
dc.subject (關鍵詞) abstract interpretation; adversarial training; logic rule; robustness; neural network
dc.title (題名) Controllable Robustness Training
dc.type (資料類型) article
dc.identifier.doi (DOI) 10.6688/JISE.202505_41(3).0014
dc.doi.uri (DOI) https://doi.org/10.6688/JISE.202505_41(3).0014