Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Learning for Molecular Graphs via Ensemble Consensus
Rasmus Hannibal Tirsgaard, Laurits Fredsgaard, Marisa Wodrich, Mikkel Jordahn, Mikkel N. Schmidt
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 ensemble consensus loss on unlabeled data | PaiNN individual member · QM9 U0 | MAE (meV): 24.8752 -> 19.9642 | Improved |
| add ensemble consensus loss on unlabeled data | PaiNN ensemble M=4 · QM9 epsilon HOMO | MAE (meV): 76.4713 -> 73.0755 | Improved |
| use a consensus-trained individual member instead of a supervised ensemble | PaiNN · QM9 U0 | MAE (meV): 20.9101 -> 19.9642 | Improved |
| use ensemble consensus instead of mean-teacher | PaiNN ensemble · QM9 U0 | MAE (meV): 24.7288 -> 19.3816 | Improved |
| use ensemble consensus instead of PSEUDsigma | PaiNN · QM9 <R2> | MAE: 0.9099 -> 0.5605 | Improved |
| increase ensemble size from 2 to 4 with consensus training (ablation) | PaiNN ensemble · QM9 U0 | MAE: 20.43 -> 19.38 | Improved |
| add consensus loss to GCN | GCN individual · ZINC | MAE: .3163 -> .2406 | Improved |
| add consensus loss to GatedGCN | GatedGCN individual · ogbg-molhiv | AUROC: .7312 -> .7361 | No clear effect |
| add consensus loss to GIN | GIN individual · Peptides-struct | MAE: .2966 -> .2944 | No clear effect |
| add mean-teacher instead of ensemble consensus | GCN individual · ZINC | MAE: .2406 (consensus) vs .2884 (mean teacher) | Hurt |
| add consensus coupling in CIFAR-10 ResNet-18 ensemble | ResNet-18 ensemble · CIFAR-10 ensemble calibration | ECE and Brier | Hurt |
| add consensus coupling to the ensemble | individual members on ogbg-molhiv · ogbg-molhiv | calibration | Improved |
Closest papers by meaning
- Molecular Data Programming: Towards Molecule Pseudo-labeling with Systematic Weak Supervision (CVPR 2024 · similarity 0.87)
- Learning Over Molecular Conformer Ensembles: Datasets and Benchmarks (ICLR 2024 · similarity 0.87)
- Empowering Active Learning for 3D Molecular Graphs with Geometric Graph Isomorphism (NeurIPS 2024 · similarity 0.86)
- Robust Heterogeneous Graph Classification for Molecular Property Prediction with Information Bottleneck (AAAI 2025 · similarity 0.86)
- Physical Consistency Bridges Heterogeneous Data in Molecular Multi-Task Learning (NeurIPS 2024 · similarity 0.86)
- Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks (ICML 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.
