Research map / Architectures and efficiency / Pruning and efficient tuning
Efficient Training of Large Vision Models via Advanced Automated Progressive Learning
Changlin Li, Jiawei Zhang, Sihao Lin, Zongxin Yang, Junwei Liang, Xiaodan Liang, Xiaojun Chang
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 |
|---|---|---|---|
| replace manual uniform linear growth (Prog) with AutoProg-One search | DeiT-S · ImageNet-1K (100 epochs) | Top-1 accuracy (%): 72.6 -> 74.4 | Improved |
| add AutoProg-One to ViT training | VOLO-D1 · ImageNet-1K (100 epochs) | Top-1 accuracy (%): 82.6 -> 82.8 with +65.6% speedup | Improved |
| use smaller initial grow space 0.4 Omega | AutoProg-One VOLO-D1 · ImageNet-1K (100 epochs) | training speedup: +65.6% -> +85.1% (Top-1 82.8 -> 82.7) | Improved |
| add AutoProg-One (300 epochs) | DeiT-S · ImageNet-1K (300 epochs) | Top-1 accuracy (%): 79.8 -> 79.8 with +42.0% speedup | No clear effect |
| use Interpolation growth instead of RandInit (ablation) | Prog ViT · ImageNet-1K | Top-1 accuracy (%): 80.02 -> 81.78 | Improved |
| use Interpolation growth instead of Stacking (ablation) | Prog ViT · ImageNet-1K | Top-1 accuracy (%): 81.55 -> 81.78 | Improved |
| use function-preserving Identity growth instead of Interpolation (ablation) | Prog ViT · ImageNet-1K | Top-1 accuracy (%): 81.78 -> 79.32 | Hurt |
| add MoGrow (momentum growth) to Interpolation (ablation) | Prog ViT · ImageNet-1K | Top-1 accuracy (%): 81.78 -> 81.90 | Improved |
| add MoGrow to AutoProg-One (ablation) | AutoProg-One VOLO-D1 · ImageNet-1K | Top-1@Growth (%): 59.41 -> 62.14 (Top-1 82.6 -> 82.8) | Improved |
| add weight recycling to AutoProg-One (ablation) | AutoProg-One VOLO-D1 · ImageNet-1K | training speedup: 53.3% -> 65.6% (Top-1 82.8 -> 82.8) | Improved |
| add Unique Stage Identifier (embedding) to AutoProg-Zero (ablation) | AutoProg-Zero DiT · Food (FID) | FID: 8.61 -> 7.70 | Improved |
| add Unique Stage Identifier (text) to AutoProg-Zero (ablation) | AutoProg-Zero SD · CUB (FID) | FID: 8.87 -> 8.74 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DiffFit — DiffFit: Unlocking Transferability of Large Diffusion Models via Simple Parameter-efficient Fine-Tuning (ICCV 2023) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LaVin-DiT: Large Vision Diffusion Transformer (CVPR 2025 · similarity 0.87)
- Scalable Pre-training of Large Autoregressive Image Models (ICML 2024 · similarity 0.87)
- LMD: Faster Image Reconstruction with Latent Masking Diffusion (AAAI 2024 · similarity 0.86)
- Beyond Random Augmentations: Pretraining with Hard Views (ICLR 2025 · similarity 0.86)
- Data-efficient Large Vision Models through Sequential Autoregression (ICML 2024 · similarity 0.85)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.85)
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.
