Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
MARS: Meta-learning as Score Matching in the Function Space
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace parametric score matching with SSGE estimates (ablation) | MARS-GP · SwissFEL, Physionet-GCS, Physionet-HCT, Berkeley-Sensor, and Argus-Control | RMSE: SwissFEL: 0.391 $\pm$ 0.011 -> 0.449 $\pm$ 0.027; Physionet-GCS: 1.471 $\pm$ 0.083 -> 3.292 $\pm$ 0.562 | Hurt |
| Add spectral normalization to the score network (ablation) | MARS-GP · SwissFEL, Physionet-GCS, Physionet-HCT, Berkeley-Sensor, and Argus-Control | calibration error: SwissFEL: 0.233 $\pm$ 0.041 -> 0.035 $\pm$ 0.002; Berkeley-Sensor: 0.192 $\pm$ 0.018 -> 0.080 $\pm$ 0.005 | Helped |
| Sample from GP posterior marginals instead of using posterior means (ablation) | MARS-GP · SwissFEL, Physionet-GCS, Physionet-HCT, Berkeley-Sensor, and Argus-Control | RMSE: Physionet-HCT: 5.995 $\pm$ 1.108 -> 2.309 $\pm$ 0.041; Argus-Control: 0.073 $\pm$ 0.003 -> 0.013 $\pm$ 0.001 | Helped |
| Use MARS-GP rather than the strongest listed baseline | Meta-learning regression · Physionet-HCT | RMSE: 2.361 $\pm$ 0.047 -> 2.309 $\pm$ 0.041 | Helped |
| Use MARS-GP rather than the strongest listed baseline | Meta-learning regression · Berkeley-Sensor | RMSE: 0.121 $\pm$ 0.027 -> 0.116 $\pm$ 0.024 | Helped |
| Use MARS-GP rather than the strongest listed baseline | Meta-learning regression · Argus-Control | RMSE: 0.016 $\pm$ 0.005 -> 0.013 $\pm$ 0.001 | Helped |
Closest papers by meaning
- Meta-Learning Reliable Priors in the Function Space (NeurIPS 2021 · similarity 0.92)
- More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms (ICML 2024 · similarity 0.89)
- Recasting Gradient-Based Meta-Learning as Hierarchical Bayes (ICLR 2018 · similarity 0.89)
- PACOH: Bayes-Optimal Meta-Learning with PAC-Guarantees (ICML 2021 · similarity 0.88)
- MALIBO: Meta-learning for Likelihood-free Bayesian Optimization (ICML 2024 · similarity 0.87)
- Towards Automated Knowledge Integration From Human-Interpretable Representations (ICLR 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.