Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Robustness of Deep-Learning-Based Image Reconstruction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train with a learned, measurement-space adversarial perturbation generator in a min-max objective | Deep CS reconstruction networks · MNIST and CelebA with Gaussian measurement matrices | worst-case reconstruction error ρ̂ | Helped |
| Train with a learned, measurement-space adversarial perturbation generator in a min-max objective | Deep CS reconstruction networks · MNIST and CelebA with DCT measurement matrices | worst-case reconstruction error ρ̂ | Helped |
| Use adversarial min-max training rather than standard training | Deep CS reconstruction networks · Gaussian measurements, MNIST and CelebA, ε = 0 | reconstruction loss | Helped |
| Use adversarial min-max training rather than standard training | Deep CS reconstruction networks · DCT measurements, MNIST and CelebA, ε = 0 | worst-case reconstruction error ρ̂ | No clear effect |
| Use adversarial min-max training rather than standard or Parseval training | Deep CS reconstruction networks · DCT measurements, MNIST and CelebA, increasing ε | worst-case reconstruction error ρ̂ | Helped |
| Use adversarial min-max training rather than standard or other regularized training | Deep CS reconstruction networks · Gaussian measurements, MNIST and CelebA, varying ε | worst-case reconstruction error ρ̂ | Helped |
| Replace well-conditioned Gaussian measurements with relatively ill-conditioned DCT measurements (ablation) | Deep CS reconstruction networks · MNIST and CelebA, increasing ε | gap in ρ̂ between adversarial and other training schemes | Mixed |
| Use the min-max objective instead of an exact inverse | Linear reconstruction with an ill-conditioned measurement matrix · Linear reconstruction, adversarial measurement perturbations | sensitivity to adversarial perturbations | Helped |
| Learn the linear reconstruction with the min-max scheme | Linear network · Simulated linear reconstruction | distance to the theoretically derived solution: 0.024; 0.034 | Helped |
| Increase measurement perturbation size ε (ablation) | Adversarially trained and baseline reconstruction networks · MNIST and CelebA, Gaussian and DCT measurements | relative robustness advantage | Mixed |
| Add a learned generator to produce sample-dependent measurement perturbations instead of using a common perturbation | Adversarial training for image reconstruction · Deep CS reconstruction | robustness | Helped |
Closest papers by meaning
- UGoDIT: Unsupervised Group Deep Image Prior Via Transferable Weights (NeurIPS 2025 · similarity 0.86)
- Weakly Convex Regularisers for Inverse Problems: Convergence of Critical Points and Primal-Dual Optimisation (ICML 2024 · similarity 0.86)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.86)
- NPN: Non-Linear Projections of the Null-Space for Imaging Inverse Problems (NeurIPS 2025 · similarity 0.86)
- A Unified Model for Compressed Sensing MRI Across Undersampling Patterns (CVPR 2025 · similarity 0.86)
- Robustness of Deep Learning for Accelerated MRI: Benefits of Diverse Training Data (ICML 2024 · similarity 0.86)
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.