Research map / Architectures and efficiency / Attention and transformers
Emerging Property of Masked Token for Effective Pre-training
Hyesong Choi, Hunsang Lee, Seyoung Joung, Hyejin Park, Jiyeong Kim, 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 masked token optimization to SimMIM and BootMAE pre-training | SimMIM, BootMAE · ImageNet-1K ViT-B pre-training epochs | pre-training epochs to reach 800-epoch performance: 54% and 58% reduction | Improved |
| add MTO to MAE and ConMIM pre-training | MAE, ConMIM · ImageNet-1K ViT-B | pre-training epochs: 50% reduction | Improved |
| add MTO to MAE with ViT-L | MAE ViT-L · ImageNet-1K | pre-training epochs: 43% reduction | Improved |
| use ranking loss alone without sparsity or entropy losses (ablation) | SimMIM + MTO · ImageNet-1K 400 epochs | Top-1 accuracy: 82.8 vs 83.2-83.3 for single other losses | Hurt |
| add entropy loss L_e to ranking loss (ablation) | SimMIM + MTO · ImageNet-1K 400 epochs | Top-1 accuracy: 82.8 -> 83.4 | Improved |
| add all three losses instead of two (ablation) | SimMIM + MTO · ImageNet-1K 400 epochs | Top-1 accuracy: 83.4 -> 83.5 | Improved |
| add sparsity loss to SimMIM (ablation) | SimMIM + MTO · ImageNet-1K 400 epochs | Top-1 accuracy: 83.2 as sole loss | Improved |
Closest papers by meaning
- On the Role of Discrete Tokenization in Visual Representation Learning (ICLR 2024 · similarity 0.91)
- Morphing Tokens Draw Strong Masked Image Models (ICLR 2025 · similarity 0.90)
- Representation Deficiency in Masked Language Modeling (ICLR 2024 · similarity 0.88)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.88)
- Learning Mask Invariant Mutual Information for Masked Image Modeling (ICLR 2025 · similarity 0.88)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.88)
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.
