Research map / Optimization / Optimizers
Hierarchical Optimization-Derived Learning
Risheng Liu, Xuan Liu, Shangzhi Zeng, Jin Zhang, Yixuan Zhang
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 HODL hierarchical training instead of ENA (5 layers) | ENA · Set14 constrained sparse coding | PSNR: 15.59 -> 18.82 | Improved |
| use HODL instead of UNH (5 layers) | UNH · Set14 constrained sparse coding | PSNR: 10.47 -> 18.82 | Improved |
| increase layers from 5 to 25 (ablation) | HODL · Set14 constrained sparse coding | PSNR: 18.82 -> 18.98 | No clear effect |
| increase layers from 5 to 25 (ablation) | ENA · Set14 constrained sparse coding | PSNR: 15.59 -> 15.64 | No clear effect |
| increase layers from 5 to 25 (ablation) | UNH · Set14 constrained sparse coding | PSNR: 10.47 -> 11.31 | Improved |
| use ALM instead of ADMM as base solver (ablation) | ENA · Set14 constrained sparse coding | PSNR: 15.58 -> 13.78 | Hurt |
| use ALM instead of ADMM as base solver (ablation) | UNH · Set14 constrained sparse coding | PSNR: 11.27 -> 7.32 | Hurt |
| HODL with ALM versus ENA with ADMM | HODL · Set14 constrained sparse coding | PSNR: 15.58 -> 18.64 | Improved |
| add non-expansive property of the operator (ablation) | HODL · constrained sparse coding convergence | gradient of learning objective: reduces the gradient of the learning objective by an order of magnitude | Improved |
| use aggregation (mu greater than 0) instead of sHODL (mu equals 0) (ablation) | aHODL · constrained sparse coding convergence | convergence of optimization variable | Mixed |
| drop aggregation to use simplified sHODL (ablation) | aHODL · constrained sparse coding | computational cost | Improved |
| use HODL instead of RCDNet on Rain100L | RCDNet · Rain100L | PSNR: 40.00 -> 40.07 | No clear effect |
Closest papers by meaning
- Optimization-Derived Learning with Essential Convergence Analysis of Training and Hyper-training (ICML 2022 · similarity 0.87)
- A Stochastic Approach to Bi-Level Optimization for Hyperparameter Optimization and Meta Learning (AAAI 2025 · similarity 0.83)
- Linearly Decomposing and Recomposing Vision Transformers for Diverse-Scale Models (NeurIPS 2024 · similarity 0.83)
- Prodigy: An Expeditiously Adaptive Parameter-Free Learner (ICML 2024 · similarity 0.82)
- Delve into the Applicability of Advanced Optimizers for Multi-Task Learning (NeurIPS 2026 · similarity 0.82)
- Decouple to Generalize: Context-First Self-Evolving Learning for Data-Scarce Vision-Language Reasoning (CVPR 2026 · similarity 0.82)
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.
