Research map / Graph learning / Graph self-supervised learning
PRODIGY: Enabling In-context Learning Over Graphs
Qian Huang, Hongyu Ren, Peng Chen, Gregor Kržmanc, Daniel Zeng, Percy Liang, Jure Leskovec
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 |
|---|---|---|---|
| Remove attribute prediction loss (ablation) | PG-NM · arXiv 3-way | accuracy (%): 65.74 -> 72.50 | Improved |
| Remove attribute prediction loss (ablation) | PG-NM · arXiv 40-way | accuracy (%): 18.03 -> 23.69 | Improved |
| Remove DropNode and ZeroNode augmentations (ablation) | PG-NM · arXiv 3-way | accuracy (%): 68.98 -> 72.50 | Improved |
| Remove both attribute prediction and augmentations (ablation) | PG-NM · arXiv 3-way | accuracy (%): 66.53 -> 72.50 | Improved |
| Remove attribute prediction, augmentations and task graph GNN (ablation) | PG-NM · arXiv 3-way | accuracy (%): 63.60 -> 72.50 | Improved |
| Reduce neighbor-matching prompt shots from 3 to 1 (ablation) | PG-NM · arXiv 3-way | accuracy (%): 69.13 -> 72.50 | Improved |
| Combine neighbor matching and multi-task pretraining (full PRODIGY vs PG-MT) | PG-MT · arXiv 3-way | accuracy (%): 65.64 -> 73.09 | Improved |
| Combine neighbor matching and multi-task pretraining (full PRODIGY vs PG-NM) | PG-NM · arXiv 40-way | accuracy (%): 23.69 -> 25.13 | Improved |
| Combine both tasks (full PRODIGY vs PG-MT) | PG-MT · FB15K-237 5-way | accuracy (%): 89.15 -> 88.02 | Mixed |
| Add contrastive pretraining baseline replaced by PRODIGY | Contrastive · arXiv 5-way | accuracy (%): 51.63 -> 61.52 | Improved |
| Pretrain with PRODIGY vs pretrain plus finetune head | Finetune · arXiv 40-way | accuracy (%): 8.04 -> 25.13 | Improved |
| Compare self-supervised PG-NM to meta-learning TENT | TENT · arXiv 3-way test labels | accuracy (%): 65.13 -> 69.07 | Improved |
Later papers that compared with this method
4 later accepted papers report Prodigy as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Modality-free Graph In-context Alignment (ICLR 2026 · better on 6 of 6 shared results)
- Hard-Constrained Graph Generation with Discrete-Projection Diffusion (ICML 2026 · better on 3 of 4 shared results)
- Context-aware Graph Meta-learning (AAAI 2026 · better on 1 of 1 shared result)
- One For All: Towards Training One Graph Model For All Classification Tasks (ICLR 2024 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- One For All: Towards Training One Graph Model For All Classification Tasks (ICLR 2024 · similarity 0.90)
- ICLR: In-Context Learning of Representations (ICLR 2025 · similarity 0.90)
- One Prompt Fits All: Universal Graph Adaptation for Pretrained Models (NeurIPS 2025 · similarity 0.88)
- GRAVER: Generative Graph Vocabularies for Robust Graph Foundation Models Fine-tuning (NeurIPS 2025 · similarity 0.87)
- ProG: A Graph Prompt Learning Benchmark (NeurIPS 2024 · similarity 0.87)
- Towards Graph Foundation Models: Learning Generalities Across Graphs via Task-Trees (ICML 2025 · similarity 0.87)
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.
