Research map / Probabilistic and scientific modeling / Molecules and proteins
Generative Augmented Inference
Cheng Lu, Mengxin Wang, Dennis Zhang, Heng Zhang
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 |
|---|---|---|---|
| treat AI outputs as features in a Neyman-orthogonal score (GAI) instead of surrogate labels (PPI) | GAI · Vaccine conjoint (nP=50) | MAPE (%): 32.02 -> 16.86 (Primary vs GAI) | Improved |
| use GAI rather than PPI with low-accuracy (54%) LLM labels | GAI vs PPI · Vaccine conjoint (nP=100) | MAPE (%): 44.57 -> 17.52 | Improved |
| naively pool human and AI labels without correction | Primary estimator · Vaccine conjoint | 95% CI coverage (%): 88.00 -> 27.09 (nP=200) | Hurt |
| use GAI instead of PPI for CI width in small primary samples | GAI vs PPI · Vaccine conjoint (nP=50) | 95% CI width: 9.28 -> 1.94 | Improved |
| use high-dimensional text embeddings of CoT as auxiliary z (GAI Emb) instead of discrete labels (ablation) | GAI · Vaccine conjoint | MAPE (%) / CI coverage: MAPE 16.64 -> 16.24 and coverage 90.91 -> 94.55 (nP=200), CI width 0.96 -> 1.05 | Mixed |
| use GAI instead of PPI-family methods with well-calibrated continuous AI probabilities | GAI · Census health insurance (nP=1000) | MAPE (%): 291.80 -> 141.49 (Primary vs GAI) | Improved |
| GAI vs PPI on coverage with calibrated AI predictions | GAI · Census health insurance | 95% CI coverage (%): 94.00 -> 99.00 (nP=1000) | Improved |
| GAI vs PPI on CI width with calibrated AI predictions | GAI · Census health insurance | 95% CI width: 0.96 -> 0.93 (nP=1000) | No clear effect |
| GAI vs other benchmarks coverage on conjoint with nP=200 | GAI · Vaccine conjoint (nP=200) | 95% CI coverage (%): PPI 92.91 vs GAI 90.91 | Mixed |
Closest papers by meaning
- Prediction-Powered Inference Across Many Tasks for AI Evaluations and Social Science Research (NeurIPS 2026 · similarity 0.87)
- Regression for the Mean: Auto-Evaluation and Inference with Few Labels through Post-hoc Regression (ICML 2025 · similarity 0.86)
- Stratified Prediction-Powered Inference for Effective Hybrid Evaluation of Language Models (NeurIPS 2024 · similarity 0.86)
- Maximizing Signal in Human-Model Preference Alignment (AAAI 2025 · similarity 0.85)
- AHA: Human-Assisted Out-of-Distribution Generalization and Detection (NeurIPS 2024 · similarity 0.85)
- Generative Forests (NeurIPS 2024 · 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.
