Research map / Learning theory and algorithms / Combinatorial and search optimization
Parametric Majorization for Data-Driven Energy Minimization Methods
Jonas Geiping, Michael Moeller
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 |
|---|---|---|---|
| add the iterative majorization scheme on top of the single Bregman-type surrogate for 3 3x3 filters (ablation) | parametric majorization (analysis operator denoising) · BSDS denoising (3 3x3 filters) | PSNR: 26.66 -> 27.66 | Improved |
| add the iterative majorization scheme for 48 7x7 filters (ablation) | parametric majorization (analysis operator denoising) · BSDS denoising (48 7x7 filters) | PSNR: 27.41 -> 28.03 | Improved |
| add the iterative majorization scheme for 96 9x9 filters (ablation) | parametric majorization (analysis operator denoising) · BSDS denoising (96 9x9 filters) | PSNR: 27.46 -> 28.03 | Improved |
| increase filter bank from 3 3x3 to 48 7x7 filters (single surrogate) | parametric majorization (analysis operator denoising) · BSDS denoising | PSNR: 26.66 -> 27.41 | Improved |
| increase filters from 48 7x7 to 96 9x9 (single surrogate) | parametric majorization (analysis operator denoising) · BSDS denoising | PSNR: 27.41 -> 27.46 | No clear effect |
| scale model from 48 7x7 to 96 9x9 filters in training time | parametric majorization (analysis operator denoising) · BSDS denoising | training time per surrogate: 02:45 -> 01:43 | Improved |
| learned model with surrogate instead of prior implicit-differentiation training for 96 9x9 filters | analysis operator denoising · BSDS denoising | training time: 20 days -> 01:43 | Improved |
| add learned linear correction term trained with gradient-penalty surrogate to Huber-TV CT reconstruction | Huber-regularized CT reconstruction · Shepp-Logan phantom CT | PSNR: almost 2dB | Improved |
| add perimeter (TV) term trained end-to-end with surrogate to linear segmentation model | variational segmentation, linear model · Cityscapes subset (4 images) | training accuracy | Improved |
| add a small number of iterations of the iterative scheme (four) to the surrogate for segmentation (ablation) | variational segmentation, linear model · Cityscapes subset (4 images) | training accuracy | Improved |
Closest papers by meaning
- Complexity of Discrete Energy Minimization Problems (ECCV 2016 · similarity 0.87)
- Controlling Neural Networks via Energy Dissipation (ICCV 2019 · similarity 0.86)
- Designing Energy-Efficient Convolutional Neural Networks Using Energy-Aware Pruning (CVPR 2017 · similarity 0.85)
- Joint Learning of Energy-based Models and their Partition Function (ICML 2025 · similarity 0.84)
- A Stochastic Approach to Bi-Level Optimization for Hyperparameter Optimization and Meta Learning (AAAI 2025 · similarity 0.83)
- Learning normalized image densities via dual score matching (NeurIPS 2025 · similarity 0.83)
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.
