ML Security PapersDDSA: Dual-Domain Strategic Attack for Spatial-Temporal Efficiency in Adversarial Robustness Testing
Jinwei Hu 1, Shiyuan Meng 2, Yi Dong 1, Xiaowei Huang 1
Published on arXiv
2601.14302
Input Manipulation Attack
OWASP ML Top 10 — ML01
Key Finding
Dual-domain (temporal + spatial) attack strategy achieves substantial computational resource conservation while maintaining adversarial attack effectiveness on priority object classes in resource-constrained real-time applications.
DDSA (Dual-Domain Strategic Attack)
Novel technique introduced
Image transmission and processing systems in resource-critical applications face significant challenges from adversarial perturbations that compromise mission-specific object classification. Current robustness testing methods require excessive computational resources through exhaustive frame-by-frame processing and full-image perturbations, proving impractical for large-scale deployments where massive image streams demand immediate processing. This paper presents DDSA (Dual-Domain Strategic Attack), a resource-efficient adversarial robustness testing framework that optimizes testing through temporal selectivity and spatial precision. We introduce a scenario-aware trigger function that identifies critical frames requiring robustness evaluation based on class priority and model uncertainty, and employ explainable AI techniques to locate influential pixel regions for targeted perturbation. Our dual-domain approach achieves substantial temporal-spatial resource conservation while maintaining attack effectiveness. The framework enables practical deployment of comprehensive adversarial robustness testing in resource-constrained real-time applications where computational efficiency directly impacts mission success.
Key Contributions
- Scenario-aware temporal trigger function that selects critical frames for adversarial testing based on class priority and model uncertainty, reducing unnecessary per-frame overhead
- Explainable AI-guided spatial targeting using Integrated Gradients to locate influential pixel regions for focused perturbation instead of full-image attacks
- DDSA framework combining temporal selectivity and spatial precision for resource-efficient adversarial robustness testing in large-scale, real-time image processing deployments
🛡️ Threat Analysis
DDSA generates adversarial perturbations (via FGSM/PGD) that cause misclassification at inference time. The dual-domain contribution — temporal frame selection (when to attack) and XAI-guided pixel targeting via Integrated Gradients (where to attack) — is an optimization on top of gradient-based adversarial example generation. A framework for generating adversarial examples is an attack, not a benchmark.