Research map / Architectures and efficiency / Attention and transformers
Salience-Based Adaptive Masking: Revisiting Token Dynamics for Enhanced Pre-training
Hyesong Choi, Hyejin Park, Kwang Moo Yi, Sungmin Cha, Dongbo Min
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 SBAM salience-based masking to MAE ViT-L | MAE ViT-L · ImageNet-1K fine-tuning | top-1 accuracy (%): 84.3 -> 85.1 | Improved |
| add SBAM to iBoT ViT-B (100 epochs) | iBoT · ImageNet-1K fine-tuning | top-1 accuracy (%): 71.5 -> 74.4 | Improved |
| add SBAM to BootMAE and CMAE | BootMAE, CMAE · ImageNet-1K fine-tuning | top-1 accuracy (%): 84.1 -> 84.8 (BootMAE), 83.8 -> 84.5 (CMAE) | Improved |
| outgoing-weight token salience instead of attention-based masking (AMT) | MAE ViT-B · ImageNet-1K fine-tuning | accuracy (%): 82.8 -> 83.6 | Improved |
| SBAM instead of AttMask in iBoT | iBoT ViT-B · ImageNet-1K fine-tuning | accuracy (%): 72.8 -> 74.4 | Improved |
| add random noise to token salience when masking (ablation) | SBAM · ImageNet-1K ViT-L MAE 400 epochs | accuracy | Improved |
| vary delta r and delta hyperparameters of AMR (ablation) | SBAM · ImageNet-1K ViT-L | fine-tuning vs linear probing accuracy | Mixed |
| SBAM masking across masking ratios vs random masking | MAE ViT-L · ImageNet-1K | accuracy across masking ratios | Improved |
Closest papers by meaning
- Patch-Aware Sample Selection for Efficient Masked Image Modeling (AAAI 2024 · similarity 0.87)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.87)
- Morphing Tokens Draw Strong Masked Image Models (ICLR 2025 · similarity 0.86)
- Learning Mask Invariant Mutual Information for Masked Image Modeling (ICLR 2025 · similarity 0.86)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.86)
- ALTo: Adaptive-Length Tokenizer for Autoregressive Mask Generation (NeurIPS 2025 · 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.
