Research map / Optimization / Implicit bias and generalization
Convex Regularization behind Neural Reconstruction
Arda Sahiner, Morteza Mardani, Batu Ozturkler, Mert Pilanci, John M. Pauly
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 |
|---|---|---|---|
| Use convex dual optimization instead of non-convex primal training | Two-layer fully convolutional ReLU denoising network · MNIST denoising, tested Gaussian noise levels | training and test objective | No clear effect |
| Use convex dual optimization instead of non-convex primal training | Two-layer fully convolutional ReLU reconstruction network · fastMRI knee reconstruction, undersampling rates R=2,4,8 | training and test loss | No clear effect |
| Use convex dual optimization instead of non-convex primal training | Two-layer fully convolutional ReLU denoising network · MNIST denoising with exponentially distributed noise, λ=1.15 | objective | Improved |
| Increase sampled sign patterns to approximately 4000 (ablation) | Convex dual network · MNIST denoising, Gaussian noise σ=0.5 | approximation to primal objective (duality gap) | Improved |
| Use 125 sampled sign patterns (ablation) | Convex dual network · MNIST denoising, Gaussian noise σ=0.1 | approximation to primal objective (duality gap) | Improved |
Closest papers by meaning
- Weakly Convex Regularisers for Inverse Problems: Convergence of Critical Points and Primal-Dual Optimisation (ICML 2024 · similarity 0.90)
- Learning Proximal Operators: Using Denoising Networks for Regularizing Inverse Imaging Problems (ICCV 2017 · similarity 0.89)
- Fast Convex Optimization for Two-Layer ReLU Networks: Equivalent Model Classes and Cone Decompositions (ICML 2022 · similarity 0.89)
- Revealing the Structure of Deep Neural Networks via Convex Duality (ICML 2021 · similarity 0.89)
- Adversarial Regularizers in Inverse Problems (NeurIPS 2018 · similarity 0.89)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.88)
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.
