Research map / Graph learning / Graph self-supervised learning
Masked BRep Autoencoder via Hierarchical Graph Transformer
Yifei Li, Kang Wu, Wenming Wu, Xiao-Ming Fu
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 dual-resolution (3x3 and 13x13) CSMA instead of single 13x13 (ablation) | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 85.83 -> 88.75 | Improved |
| use dual resolution instead of triple resolution (7x7 added) (ablation) | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 86.41 -> 88.75 | Improved |
| add dual-resolution CSMA instead of single resolution (ablation) | Masked BRep Autoencoder · MFInstSeg (1% labels) | Acc (%): 95.19 -> 97.31 | Improved |
| use dual resolution instead of triple resolution (ablation) | Masked BRep Autoencoder · MFInstSeg (100% labels) | Acc (%): 99.42 -> 99.53 | Improved |
| add 70% masking during pre-training instead of 0% mask (ablation) | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 66.41 -> 88.75 | Improved |
| add 70% masking during pre-training (ablation) | Masked BRep Autoencoder · MFInstSeg (100% labels) | Acc (%): 99.43 -> 99.53 | No clear effect |
| Transformer before MPNN (global to local) instead of MPNN first (ablation) | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 76.39 -> 88.75 | Improved |
| mask raw inputs instead of masking gAAG features after encoder (ablation) | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 78.63 -> 88.75 | Improved |
| fine-tune the encoder instead of freezing it (ablation) | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 82.01 -> 88.75 | Improved |
| fine-tune the encoder instead of freezing it (ablation) | Masked BRep Autoencoder · MFInstSeg (100% labels) | Acc (%): 98.64 -> 99.53 | Improved |
| include SolidLetter samples in pre-training corpus | Masked BRep Autoencoder · SolidLetter classification (10-shot) | Acc (%): 81.83 -> 82.51 | Improved |
| self-supervised pre-training vs training BRepFormer from scratch | Masked BRep Autoencoder · MFInstSeg (0.1% labels) | Acc (%): 33.43 -> 88.75 | Improved |
Closest papers by meaning
- Self-supervised Masked Graph Autoencoder via Structure-aware Curriculum (ICML 2025 · similarity 0.88)
- Teacher-guided Edge Discriminator for Personalized Graph Masked Autoencoder (AAAI 2025 · similarity 0.86)
- Unifying and Enhancing Graph Transformers via a Hierarchical Mask Framework (NeurIPS 2025 · similarity 0.84)
- Linear Transformer Topological Masking with Graph Random Features (ICLR 2025 · similarity 0.84)
- The quest for the GRAph Level autoEncoder (GRALE) (NeurIPS 2025 · similarity 0.84)
- Self-Guided Masked Autoencoder (NeurIPS 2024 · similarity 0.83)
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.
