Research map / Probabilistic and scientific modeling / Molecules and proteins
General Synthetic-Powered Inference
Meshi Bashari, Yonghoon Lee, Roy Maor Lotan, Edgar Dobriban, Yaniv Romano
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 |
|---|---|---|---|
| wrap the base method with GESPI (pooled real+synthetic run plus guardrail at alpha+eps) | conformal prediction · ImageNet 30 classes (FLUX synthetic) | coverage variability | Improved |
| wrap conformal risk control with GESPI using AlphaFold MSAs as synthetic data | OnlyReal conformal risk control · CASP-14 AlphaFold2 (n=10 real proteins, N=1,000 synthetic) | abstention rate | Improved |
| add synthetic data via GESPI to FWER-controlling outlier detection | OnlyReal conformal outlier detection · Shuttle, Credit card, KDDCup99 (q=2.5%, eps=10%) | power at equal empirical error | Improved |
| use GESPI instead of Guardrail alone for outlier detection | Guardrail · Shuttle, Credit card, KDDCup99 | FWER variability | Improved |
| synthetic data follows the null while real data follows the alternative (ablation) | GESPI binomial test · simulated binomial test (n=50, N=500) | power | No clear effect |
| both real and synthetic data under the alternative (rho=0.6, rho_synt=0.55) (ablation) | GESPI binomial test · simulated binomial test (n=50, N=500) | power | Improved |
| increase eps (ablation) | GESPI binomial test · simulated, alternative holds for both datasets | power | Improved |
| synthetic data follows the alternative while real data follows the null (ablation) | GESPI binomial test · simulated binomial test | Type I error | Hurt |
| increase real dataset size n (ablation) | GESPI binomial test · simulated, alternative for both | power | Improved |
| increase synthetic dataset size N beyond 500 (ablation) | GESPI binomial test · simulated, alternative for both | power | No clear effect |
Closest papers by meaning
- Synthetic-powered predictive inference (NeurIPS 2025 · similarity 0.88)
- Anytime-valid, Bayes-assisted, Prediction-Powered Inference (NeurIPS 2025 · similarity 0.84)
- Fisher-R1: Training LLM Agents for Reliable Hypothesis Testing (NeurIPS 2026 · similarity 0.84)
- FunctionEvolve: Structure-Guided Symbolic Regression with LLMs (NeurIPS 2026 · similarity 0.84)
- The Silent Hyperparameter: Quantifying the Impact of Inference Backends on LLM Reproducibility (NeurIPS 2026 · similarity 0.84)
- Test-time Augmentation Improves Efficiency in Conformal Prediction (CVPR 2025 · similarity 0.84)
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.
