Improving Transferability of Adversarial Attacks via Frequency-Consistent Regularization
Adversarial examples have revealed the vulnerability of deep neural networks, and their transferability makes black-box attacks particularly concerning. However, perturbations crafted on a surrogate model often do not remain sufficiently effective on unseen target models. In this paper, we revisit t...
সংরক্ষণ করুন:
| প্রধান লেখক: | , , |
|---|---|
| বিন্যাস: | Artigo |
| ভাষা: | Inglês |
| প্রকাশিত: |
MDPI AG
2026-04-01
|
| মালা: | Applied Sciences |
| বিষয়গুলি: | |
| অনলাইন ব্যবহার করুন: | https://www.mdpi.com/2076-3417/16/8/3748 |
| ট্যাগগুলো: |
কোনো ট্যাগ নেই, প্রথমজন হিসাবে ট্যাগ করুন!
|