Research map / Architectures and efficiency / Pruning and efficient tuning
Neural Network Optimization Reimagined: Decoupled Techniques for Scratch and Fine-Tuning
Xin Ning, Qiankun Li, Xiaolong Huang, Qiupu Chen, He Feng, Weijun Li, Prayag Tiwari, X Y Liu
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 DualOpt_Scratch real-time layer-wise weight decay | Adam (ResNet-50) · CIFAR-100 (from scratch) | accuracy: 63.47 -> 64.67 | Improved |
| add DualOpt_Scratch | SGD (ResNet-50) · CUB-200-2011 (from scratch) | accuracy: 15.59 -> 20.19 | Improved |
| add DualOpt_Scratch | Adam (Inception) · OfficeHome (from scratch) | accuracy: 31.15 -> 42.98 | Improved |
| add DualOpt_Scratch | SGD (ResNet-50) · OfficeHome (from scratch) | accuracy: 43.81 -> 42.01 | Hurt |
| add DualOpt fine-tuning (weight rollback, layer-wise penalty, diversified decay) | Full fine-tuning (ViT-B) · PACS | accuracy: 87.79 -> 94.38 | Improved |
| add DualOpt fine-tuning | Full fine-tuning (ConvNeXt-B) · CIFAR-100 | accuracy: 92.72 -> 92.86 | No clear effect |
| add DualOpt | ViT-B from scratch · ImageNet | top-1 accuracy: 81.34 -> 82.58 | Improved |
| add DualOpt | Swin-B from scratch · ImageNet | top-1 accuracy: 83.05 -> 83.89 | Improved |
| add DualOpt | Mask R-CNN full fine-tuning · COCO2017 | Bbox mAP: 40.20 -> 41.10 | Improved |
| add DualOpt | UperNet full fine-tuning · ADE20K | mIoU: 43.65 -> 44.62 | Improved |
| add layer-wise decay rate lambda*i/n instead of uniform decay (ablation) | DualOpt_Scratch (Adam) · CIFAR-100 | accuracy: 63.47 -> 64.67 | Improved |
| use inverse layer-wise decay instead of proposed (ablation) | DualOpt_Scratch (Adam) · CIFAR-100 | accuracy: 64.67 -> 62.61 | Hurt |
Closest papers by meaning
- Why Do We Need Weight Decay in Modern Deep Learning? (NeurIPS 2024 · similarity 0.88)
- Rethinking Weight Decay for Robust Fine-Tuning of Foundation Models (NeurIPS 2024 · similarity 0.88)
- Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks (ICLR 2018 · similarity 0.88)
- On The Concurrence of Layer-wise Preconditioning Methods and Provable Feature Learning (ICML 2025 · similarity 0.87)
- Pay Attention to Small Weights (NeurIPS 2025 · similarity 0.87)
- Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation (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.
