Research map / Adversarial robustness and security / Certified and empirical robustness
Training-Free Adversarial Robustness in Computational MRI
Mahdi Saberi, Chi Zhang, Mehmet Akcakaya
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply cyclic-consistency mitigation to the attacked input | MoDL · Cor-PD knee, R=4 | PSNR: 35.14 vs SMUG 28.22 and AT 33.99 | Improved |
| Combine proposed mitigation with adversarial training | AT-trained MoDL · Cor-PD knee, R=4 | PSNR: 33.99 -> 36.57 | Improved |
| Combine proposed mitigation with SMUG | SMUG MoDL · Cor-PD knee, R=4 | PSNR: 28.22 -> 34.85 | Improved |
| Combine proposed mitigation with SMUG | SMUG MoDL · Ax-FLAIR brain, R=4 | PSNR: 29.67 -> 34.67 | Improved |
| Increase attack strength from 0.01 to 0.02 against models trained at 0.01 | AT and SMUG · Cor-PD knee | visual quality | Hurt |
| Use proposed mitigation across attack strengths without retraining | Proposed mitigation · Cor-PD knee, eps 0.01 and 0.02 | visual quality | Improved |
| Use unrolled mitigation with few unrolls against matched adaptive attack (ablation) | Alg. 1 unrolled · Cor-PD adaptive attack, T=10 | PSNR: 29.47 vs 34.34 (iterative) | Hurt |
| Run mitigation until convergence instead of unrolled (ablation) | Alg. 1 · Cor-PD adaptive attack, T=10 | PSNR: 29.47 -> 34.34 | Improved |
| Increase unroll count T of adaptive attack (ablation) | Alg. 1 iterative · Cor-PD adaptive attack | PSNR: 34.34 (T=10) -> 34.01 (T=100) | Hurt |
| Use multiple cyclic reconstruction levels in the loss (ablation) | Proposed mitigation · Cor-PD knee | reconstruction quality | Hurt |
| Use suboptimally trained network or mismatched forward operator in attack (ablation) | Proposed mitigation · Cor-PD knee | reconstruction quality | No clear effect |
| Apply mitigation to unperturbed inputs | Proposed mitigation · clean inputs | reconstruction quality | No clear effect |
Closest papers by meaning
- Solving Inverse Problems With Deep Neural Networks – Robustness Included? (TPAMI 2022 · similarity 0.88)
- An Introduction to Adversarially Robust Deep Learning (TPAMI 2023 · similarity 0.87)
- Measuring Robustness in Deep Learning Based Compressive Sensing (ICML 2021 · similarity 0.87)
- Improving Robustness of Deep-Learning-Based Image Reconstruction (ICML 2020 · similarity 0.87)
- Breaking Barriers in Physical-World Adversarial Examples: Improving Robustness and Transferability via Robust Feature (AAAI 2025 · similarity 0.86)
- Robustness of Deep Learning for Accelerated MRI: Benefits of Diverse Training Data (ICML 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.
