Research map / Adversarial robustness and security / Attacks on text and graphs
ASTrA: Adversarial Self-supervised Training with Adaptive-Attacks
Prakash Chandra Chhipa, Gautam Vashishtha, Settur Jithamanyu, Rajkumar Saini, Mubarak Shah, Marcus Liwicki
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes ASTrA, a self-supervised adversarial training method that replaces hand-crafted PGD attack strategies with adaptive, instance-specific adversarial attacks that evolve with the model's learning dynamics.
Problem
Self-supervised adversarial training (self-AT) relies on hand-crafted PGD attack strategies that do not adapt to the model's evolving learning dynamics or to instance-specific image characteristics, yielding sub-optimal robustness and poor alignment between clean and adversarial distributions.
Approach
Adaptive adversarial attacks that adjust to the evolving training dynamics of the model and to per-instance characteristics, instead of fixed hand-crafted PGD strategies, used within self-supervised adversarial training to better align clean and adversarial data distributions.
Experimental setup
Builds on: PGD adversarial training self-supervised adversarial training
Closest papers by meaning
- 3SAT: A Simple Self-Supervised Adversarial Training Framework (AAAI 2025 · similarity 0.88)
- Self-Supervised Adversarial Training via Diverse Augmented Queries and Self-Supervised Double Perturbation (NeurIPS 2024 · similarity 0.88)
- Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks (CVPR 2025 · similarity 0.86)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.86)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.85)
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
