Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.26981 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912679092813824 |
|---|---|
| author | Hou, Zhichao Gao, Weizhi Liu, Xiaorui |
| author_facet | Hou, Zhichao Gao, Weizhi Liu, Xiaorui |
| contents | This work tackles a critical challenge in AI safety research under limited compute: given a fixed computation budget, how can one maximize the strength of iterative adversarial attacks? Coarsely reducing the number of attack iterations lowers cost but substantially weakens effectiveness. To fulfill the attainable attack efficacy within a constrained budget, we propose a fine-grained control mechanism that selectively recomputes layer activations across both iteration-wise and layer-wise levels. Extensive experiments show that our method consistently outperforms existing baselines at equal cost. Moreover, when integrated into adversarial training, it attains comparable performance with only 30% of the original budget. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_26981 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Fine-Grained Iterative Adversarial Attacks with Limited Computation Budget Hou, Zhichao Gao, Weizhi Liu, Xiaorui Machine Learning Artificial Intelligence This work tackles a critical challenge in AI safety research under limited compute: given a fixed computation budget, how can one maximize the strength of iterative adversarial attacks? Coarsely reducing the number of attack iterations lowers cost but substantially weakens effectiveness. To fulfill the attainable attack efficacy within a constrained budget, we propose a fine-grained control mechanism that selectively recomputes layer activations across both iteration-wise and layer-wise levels. Extensive experiments show that our method consistently outperforms existing baselines at equal cost. Moreover, when integrated into adversarial training, it attains comparable performance with only 30% of the original budget. |
| title | Fine-Grained Iterative Adversarial Attacks with Limited Computation Budget |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2510.26981 |