Research map / Label-efficient and robust learning / Domain adaptation and generalization
Bayesian Meta-Learning with Expert Feedback for Task-Shift Adaptation through Causal Embeddings
Lotta Mäkinen, Jorge Loria, Samuel Kaski
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 |
|---|---|---|---|
| condition prior on ICP causal embeddings (expert-inferred) vs no transfer | hierarchical Bayesian meta-learner · UK Biobank J44 | AUROC: 0.791 -> 0.820 | Improved |
| condition prior on ICP causal embeddings (expert-inferred) vs no transfer | hierarchical Bayesian meta-learner · UK Biobank J45 | AUROC: 0.587 -> 0.620 | Improved |
| use expert-inferred ICP embeddings vs HBM global prior | HBM · UK Biobank I21 | AUROC: 0.701 -> 0.693 | Hurt |
| replace expert-inferred with oracle ICP embeddings (ablation) | ICP embeddings · UK Biobank I21 | AUROC: 0.693 -> 0.714 | Improved |
| replace expert-inferred with oracle ICP embeddings (ablation) | ICP embeddings · UK Biobank J44 | AUROC: 0.820 -> 0.816 | No clear effect |
| replace expert-inferred with oracle CHI2 embeddings (ablation) | CHI2 embeddings · UK Biobank J44 | AUROC: 0.730 -> 0.815 | Improved |
| replace expert-inferred with oracle MR embeddings (ablation) | MR embeddings · UK Biobank J44 | AUROC: 0.805 -> 0.766 | Hurt |
| use ICP oracle embeddings instead of MR oracle (ablation) | MR embeddings (oracle) · UK Biobank G45 | AUROC: 0.651 -> 0.673 | Improved |
| add standard meta-learning transfer (MAML) vs no transfer | no-transfer baseline · UK Biobank J45 | AUROC: 0.587 -> 0.612 | Improved |
| add DKT transfer vs no transfer | no-transfer baseline · UK Biobank I21 | AUROC: 0.704 -> 0.695 | No clear effect |
| condition prior on causal embeddings instead of global prior or correlation embeddings under increasing task shift (ablation) | global-prior Bayesian meta-learner · synthetic Experiment 1 | AUROC under task shift | Improved |
| use correlation-based embeddings at in-distribution shift (ablation) | global prior · synthetic Experiment 1 (eps_OOD=0.5) | AUROC | Improved |
Closest papers by meaning
- Learning to Balance: Bayesian Meta-Learning for Imbalanced and Out-of-distribution Tasks (ICLR 2020 · similarity 0.87)
- Reconciling meta-learning and continual learning with online mixtures of tasks (NeurIPS 2019 · similarity 0.87)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 2024 · similarity 0.87)
- A Meta-Learning Approach to Bayesian Causal Discovery (ICLR 2025 · similarity 0.86)
- Scalable PAC-Bayesian Meta-Learning via the PAC-Optimal Hyper-Posterior: From Theory to Practice (JMLR 2023 · similarity 0.86)
- Estimating Interventional Distributions with Uncertain Causal Graphs through Meta-Learning (NeurIPS 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.
