Research map / Label-efficient and robust learning / Recognition and long-tail
Synthetic Model Combination: An Instance-wise Approach to Unsupervised Ensemble Learning
Alex Chan, Mihaela van der Schaar
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 separation and connection losses to the reconstruction-only representation learning (ablation) | SMC · MNIST (ten single-digit classifiers) | OneVsRest AUROC: 0.969 -> 0.972 | Improved |
| Provide only 3 feature samples per model instead of none | SMC · MNIST | AUROC: 0.50 -> 0.83 | Improved |
| Provide all training features instead of few samples | SMC · MNIST | AUROC: 0.83 -> 0.97 | Improved |
| Add representation learning step before density estimation (ablation) | SMC · MNIST | test point coverage | Improved |
| Use instance-wise SMC instead of a global ensemble | SMC · Sin-wave regression with disjoint training domains | prediction accuracy | Improved |
| Make model training domains fully overlap | SMC · Sin-wave regression | prediction accuracy vs global ensembles | No clear effect |
| Move model training domains further apart (centred at 0 and 20) | SMC · Sin-wave regression | prediction accuracy in uncovered region | Hurt |
| Replace global model averaging with SMC | SMC · Vancomycin AUC, four observations | relative RMSE: 16.6% -> 12.8% | Improved |
| Replace global model averaging with SMC | SMC · Vancomycin AUC, a priori | relative RMSE: 41.8% -> 55.7% | Hurt |
| Increase number of concentration observations from 0 to 4 | SMC · Vancomycin AUC | relative RMSE: 41.8% -> 12.8% | Improved |
| Include models with very poor performance in the pool (high performance variability) | SMC · Vancomycin AUC a priori | relative RMSE: 32.7 -> 83.3 | Hurt |
| Increase weight of connection and separation losses within about 2.5% of reconstruction loss (ablation) | SMC networks · MNIST representation | prediction performance | Improved |
Closest papers by meaning
- Breaking the Barrier of Hard Samples: A Data-Centric Approach to Synthetic Data for Medical Tasks (ICML 2025 · similarity 0.83)
- Improving the Generation and Evaluation of Synthetic Data for Downstream Medical Causal Inference (NeurIPS 2025 · similarity 0.83)
- No Free Lunch from Random Feature Ensembles: Scaling Laws and Near-Optimality Conditions (ICML 2025 · similarity 0.83)
- AutoElicit: Using Large Language Models for Expert Prior Elicitation in Predictive Modelling (ICML 2025 · similarity 0.83)
- Blend the Separated: Mixture of Synergistic Experts for Data-Scarcity Drug-Target Interaction Prediction (AAAI 2025 · similarity 0.82)
- Independence Tests for Language Models (ICML 2025 · similarity 0.82)
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.
