Research map / Graph learning / Graph self-supervised learning
Interpretable Structure-Evolving LSTM
Xiaodan Liang, Liang Lin, Xiaohui Shen, Jiashi Feng, Shuicheng Yan, Eric P. Xing
Check 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 |
|---|---|---|---|
| Learn adaptive multi-level graph structures instead of using the fixed Graph LSTM graph | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 60.16 -> 63.57 | Improved |
| Use stochastic graph merging instead of deterministic merging at threshold 0.7 (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 62.20 -> 63.57 | Improved |
| Use deterministic merging at threshold 0.5 (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 59.63 | Hurt |
| Use deterministic merging at threshold 0.9 (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 60.99 | Hurt |
| Use predefined multi-scale superpixel maps instead of adaptively evolved graph structures (ablation) | Graph LSTM · PASCAL-Person-Part | Average IoU: 61.02 -> 63.57 | Hurt |
| Increase the structure-evolving LSTM depth from 1 to 5 layers (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 58.19 -> 63.57 | Improved |
| Use only the 1st-level graph prediction instead of combining predictions across levels (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 57.19 | Hurt |
| Use only the 2nd-level graph prediction instead of combining predictions across levels (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 61.29 | Hurt |
| Use only the 3rd-level graph prediction instead of combining predictions across levels (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 60.13 | Hurt |
| Use only the 4th-level graph prediction instead of combining predictions across levels (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 59.87 | Hurt |
| Use only the 5th-level graph prediction instead of combining predictions across levels (ablation) | Structure-evolving LSTM · PASCAL-Person-Part | Average IoU: 63.57 -> 59.23 | Hurt |
| Learn hierarchical graph structures instead of using the fixed Graph LSTM graph | Structure-evolving LSTM · Horse-Cow (Horse) | IOU: 74.75 -> 76.87 | Improved |
Closest papers by meaning
- SSTAG: Structure-Aware Self-Supervised Learning Method for Text-Attributed Graphs (NeurIPS 2025 · similarity 0.86)
- LLaGA: Large Language and Graph Assistant (ICML 2024 · similarity 0.86)
- UniGTE: Unified Graph–Text Encoding for Zero-Shot Generalization across Graph Tasks and Domains (NeurIPS 2025 · similarity 0.85)
- Multi-View Empowered Structural Graph Wordification for Language Models (AAAI 2025 · similarity 0.85)
- Harnessing Explanations: LLM-to-LM Interpreter for Enhanced Text-Attributed Graph Representation Learning (ICLR 2024 · similarity 0.85)
- GOFA: A Generative One-For-All Model for Joint Graph Language Modeling (ICLR 2025 · similarity 0.85)
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.
