Research map / Label-efficient and robust learning / Self-supervised and contrastive
Modality-Agnostic Self-Supervised Learning with Meta-Learned Masked Auto-Encoder
Huiwon Jang, Jihoon Tack, Daewon Choi, Jongheon Jeong, Jinwoo Shin
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 MetaMAE instead of MAE | modality-agnostic SSL · Genomics | accuracy: 53.6 -> 69.4 | Improved |
| Use MetaMAE instead of the strongest baseline | modality-agnostic SSL · PAMAP2 | accuracy: 85.3 -> 89.3 | Improved |
| Use MetaMAE instead of the strongest baseline | modality-agnostic SSL pretrained on Pfam · Secondary Structure | accuracy: 67.6 -> 65.9 | Hurt |
| Add a Transformer decoder (ablation) | MAE · PAMAP2 | accuracy: 85.3 -> 86.5 | Improved |
| Add gradient-based latent adaptation (ablation) | MAE with decoder · PAMAP2 | accuracy: 86.5 -> 88.3 | Improved |
| Add task contrastive learning (ablation) | MAE with decoder and gradient adaptation · LibriSpeech | accuracy: 64.5 -> 79.8 | Improved |
| Use decoder depth 2 instead of 4 (ablation) | MAE · Pfam | accuracy: 44.7 -> 61.4 | Improved |
| Use decoder depth 6 instead of 4 (ablation) | MAE · LibriSpeech | accuracy: 64.1 -> 74.1 | Improved |
| Set nearby-token ratio r to 0.1 rather than 0 (ablation) | MetaMAE · PAMAP2 | accuracy: 87.5 -> 89.3 | Improved |
| Set nearby-token ratio r to 1.0 rather than 0.1 (ablation) | MetaMAE · PAMAP2 | accuracy: 89.3 -> 84.2 | Hurt |
| Share three hyperparameters across modalities (ablation) | MetaMAE · Genomics | accuracy: 69.4 -> 55.4 | Hurt |
| Use five latent adaptation steps instead of one (ablation) | MetaMAE · PAMAP2 | accuracy: 89.3 -> 89.6 | Improved |
Closest papers by meaning
- Self-Guided Masked Autoencoder (NeurIPS 2024 · similarity 0.88)
- Self-Guided Masked Autoencoders for Domain-Agnostic Self-Supervised Learning (ICLR 2024 · similarity 0.88)
- Closest Neighbors are Harmful for Lightweight Masked Auto-encoders (CVPR 2025 · similarity 0.86)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.86)
- R-MAE: Regions Meet Masked Autoencoders (ICLR 2024 · similarity 0.86)
- Learning Mask Invariant Mutual Information for Masked Image Modeling (ICLR 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.
