Research map / Label-efficient and robust learning / Self-supervised and contrastive
Masked Autoencoders that Listen
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 shifted local attention in the decoder rather than global attention (ablation) | Audio-MAE · AudioSet-2M | mAP: 46.8 -> 47.3 | Helped |
| use shifted local attention in the decoder rather than global attention (ablation) | Audio-MAE · ESC-50 | accuracy (%): 93.6 -> 94.1 | Helped |
| use shifted local attention in the decoder rather than global attention (ablation) | Audio-MAE · SID | accuracy (%): 94.1 -> 94.8 | Helped |
| use 16 decoder layers instead of 2 (ablation) | Audio-MAE · AudioSet-2M | mAP: 46.8 -> 47.3 | Helped |
| increase decoder width from 256 to 512 (ablation) | Audio-MAE · AudioSet-2M | mAP: 46.9 -> 47.3 | Helped |
| increase decoder width from 512 to 768 (ablation) | Audio-MAE · AudioSet-2M | mAP: 47.3 -> 47.3 | No clear effect |
| use overlapping patches with stride (10,10) instead of non-overlapping patches (ablation) | Audio-MAE · AudioSet-2M | mAP: 47.3 -> 47.3 | No clear effect |
| increase pre-training data from 1% to 100% of AudioSet-2M (ablation) | Audio-MAE · AudioSet-2M | mAP: 39.4 -> 47.3 | Helped |
| extend pre-training from 24 to 40 epochs (ablation) | Audio-MAE · AudioSet-2M | mAP: 47.2 -> 47.3 | No clear effect |
| initialize from ImageNet MAE weights before AudioSet self-supervised pre-training (ablation) | Audio-MAE · AudioSet-2M | mAP: 47.3 -> 47.1 | Hurt |
| replace reconstruction loss with contrastive loss (ablation) | Audio-MAE · AudioSet-2M | mAP: 47.3 -> 46.6 | Hurt |
| add contrastive loss to reconstruction loss (ablation) | Audio-MAE · AudioSet-2M | mAP: 47.3 -> 46.8 | Hurt |
Closest papers by meaning
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.88)
- Masked Autoencoders with Multi-Window Local-Global Attention Are Better Audio Learners (ICLR 2024 · similarity 0.88)
- Self-Guided Masked Autoencoder (NeurIPS 2024 · similarity 0.87)
- CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment (CVPR 2025 · similarity 0.86)
- The Dynamic Duo of Collaborative Masking and Target for Advanced Masked Autoencoder Learning (AAAI 2025 · 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.