mlsanityData to 7 Oct 2026

On instabilities of deep learning in image reconstruction - Does AI come at a cost?

arXiv 1902.05300 ↗First posted 14 Feb 2019cs.CVcited by 211

Not on the map. Why?

Abstract (arXiv)

Deep learning, due to its unprecedented success in tasks such as image classification, has emerged as a new tool in image reconstruction with potential to change the field. In this paper we demonstrate a crucial phenomenon: deep learning typically yields unstablemethods for image reconstruction. The instabilities usually occur in several forms: (1) tiny, almost undetectable perturbations, both in the image and sampling domain, may result in severe artefacts in the reconstruction, (2) a small structural change, for example a tumour, may not be captured in the reconstructed image and (3) (a counterintuitive type of instability) more samples may yield poorer performance. Our new stability test with algorithms and easy to use software detects the instability phenomena. The test is aimed at researchers to test their networks for instabilities and for government agencies, such as the Food and Drug Administration (FDA), to secure safe use of deep learning methods.

Key facts. Cited by 211 papers in the map (data through 7 Oct 2026). First posted on arXiv 14 Feb 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 32; 2022: 56; 2023: 46; 2024: 30; 2025: 17; 2026: 30.

211 papers.

Sort:

By area:

  1. 2021Solving Inverse Problems in Medical Imaging with Score-Based Generative Modelscited by 514
  2. 2021Robust Compressed Sensing MRI with Deep Generative Priorscited by 284
  3. 2021NeRP: Implicit Neural Representation Learning with Prior Embedding for Sparsely Sampled Image Reconstructioncited by 145
  4. 2021Deep Equilibrium Architectures for Inverse Problems in Imagingcited by 126
  5. 2021Bayesian imaging using Plug & Play priors: when Langevin meets Tweediecited by 96
  6. 2023Implicit Neural Representation in Medical Imaging: A Comparative Surveycited by 88
  7. 2023Image Denoising: The Deep Learning Revolution and Beyond -- A Survey Paper --cited by 71
  8. 2022Diffusion Model Based Posterior Sampling for Noisy Linear Inverse Problemscited by 58
  9. 2021Can stable and accurate neural networks be computed? -- On the barriers of deep learning and Smale's 18th problemcited by 53
  10. 2022Physics-Driven Deep Learning for Computational Magnetic Resonance Imagingcited by 49
  11. 2022Learned reconstruction methods with convergence guaranteescited by 46
  12. 2022A Neural-Network-Based Convex Regularizer for Inverse Problemscited by 40
38 more
  1. 2022How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspectivecited by 39
  2. 2021AI-based Reconstruction for Fast MRI -- A Systematic Review and Meta-analysiscited by 38
  3. 2022Distilling Robust and Non-Robust Features in Adversarial Examples by Information Bottleneckcited by 30
  4. 2022Improving Lipschitz-Constrained Neural Networks by Learning Activation Functionscited by 23
  5. 2022Theoretical Perspectives on Deep Learning Methods in Inverse Problemscited by 22
  6. 2023Score-Based Generative Models for PET Image Reconstructioncited by 22
  7. 2023IMJENSE: Scan-specific Implicit Representation for Joint Coil Sensitivity and Image Estimation in Parallel MRIcited by 22
  8. 2022Stable Deep MRI Reconstruction using Generative Priorscited by 21
  9. 2022Unsupervised Learning From Incomplete Measurements for Inverse Problemscited by 20
  10. 2021Deep J-Sense: Accelerated MRI Reconstruction via Unrolled Alternating Optimizationcited by 19
  11. 2022Do Deep Neural Networks Contribute to Multivariate Time Series Anomaly Detection?cited by 19
  12. 2022Reverse Engineering of Imperceptible Adversarial Image Perturbationscited by 17
  13. 2021The mathematics of adversarial attacks in AI -- Why deep learning is unstable despite the existence of stable neural networkscited by 16
  14. 2021An Educated Warm Start For Deep Image Prior-Based Micro CT Reconstructioncited by 16
  15. 2022Quantized Compressed Sensing with Score-based Generative Modelscited by 15
  16. 2023SMRD: SURE-based Robust MRI Reconstruction with Diffusion Modelscited by 15
  17. 2023One for Multiple: Physics-informed Synthetic Data Boosts Generalizable Deep Learning for Fast MRI Reconstructioncited by 14
  18. 2021Trusted Artificial Intelligence: Towards Certification of Machine Learning Applicationscited by 13
  19. 2022Near-Exact Recovery for Tomographic Inverse Problems via Deep Learningcited by 13
  20. 2022Reasons for the Superiority of Stochastic Estimators over Deterministic Ones: Robustness, Consistency and Perceptual Qualitycited by 13
  21. 2024DiffusionBlend: Learning 3D Image Prior through Position-aware Diffusion Score Blending for 3D Computed Tomography Reconstructioncited by 13
  22. 2021The Feasibility and Inevitability of Stealth Attackscited by 12
  23. 2023Conditional Generative Models are Provably Robust: Pointwise Guarantees for Bayesian Inverse Problemscited by 12
  24. 2025Learning Regularization Functionals for Inverse Problems: A Comparative Studycited by 12
  25. 2022Adversarial Robustness of MR Image Reconstruction under Realistic Perturbationscited by 11
  26. 2023Certified Zeroth-order Black-Box Defense with Robust UNet Denoisercited by 11
  27. 2021Interpolated Joint Space Adversarial Training for Robust and Generalizable Defensescited by 10
  28. 2022Uncertainty Estimation for Computed Tomography with a Linearised Deep Image Priorcited by 10
  29. 2023Robustness of Deep Learning for Accelerated MRI: Benefits of Diverse Training Datacited by 10
  30. 2022Ultra Low-Parameter Denoising: Trainable Bilateral Filter Layers in Computed Tomographycited by 9
  31. 2022Limitations of Deep Learning for Inverse Problems on Digital Hardwarecited by 9
  32. 2022Bayesian Inversion for Nonlinear Imaging Models using Deep Generative Priorscited by 9
  33. 2023ChatGPT is on the Horizon: Could a Large Language Model be Suitable for Intelligent Traffic Safety Research and Applications?cited by 9
  34. 2022Self-supervised Deep Unrolled Reconstruction Using Regularization by Denoisingcited by 8
  35. 2022To be or not to be stable, that is the question: understanding neural networks for inverse problemscited by 8
  36. 2023Learning Provably Robust Estimators for Inverse Problems via Jitteringcited by 8
  37. 2023Unsupervised reconstruction of accelerated cardiac cine MRI using Neural Fieldscited by 8
  38. 2024Occam's Razor for Self Supervised Learning: What is Sufficient to Learn Good Representations?cited by 8

50 most cited of 211 All 211 (JSON)