Research map / Probabilistic and scientific modeling / Causal discovery
Unveiling Prior-Data Fitted Networks on Causal Effect Estimation: Pre-Training or Fine-Tuning?
Haotian Wang, Xinpeng Lv, Hao Zou, Yanghao Xiao, Shanzhi Gu, Yang Shi, Yunxin Mao, Yuanxing Zhang, Mingyang Geng, Shaowu Yang, Haoxuan Li, Wenjing Yang and 2 more
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 |
|---|---|---|---|
| increase number of intervened variables kinterv from 2 to 7 | MSF fine-tuned TabPFN · Linear SCM | MAE: 0.3087 -> 0.3824 | Hurt |
| increase number of intervened variables kinterv from 2 to 7 | MSF fine-tuned CausalPFN · Interaction SCM | MAE: 1.3157 -> 1.6487 | Hurt |
| increase number of intervened variables kinterv from 2 to 7 | MSF fine-tuned DoPFN · Interaction SCM | MAE: 2.0135 -> 2.5897 | Hurt |
| increase number of intervened variables kinterv from 2 to 7 | MSF fine-tuned TabPFN · Nonlinear SCM | MAE: 0.1050 -> 0.1856 | Hurt |
| increase TV radius, Lipschitz factor and coverage radius | PWF · Interaction SCM | MAE: 0.425 -> 0.535 | Hurt |
| increase TV radius, Lipschitz factor and coverage radius | PWF · Linear SCM | MAE: 0.134 -> 0.158 | Hurt |
| apply point-wise interventional fine-tuning (PWF) on uncovered distribution D1 | pre-trained PFNs · Linear, Nonlinear, Interaction SCMs | MSE | Improved |
| evaluate PWF on distribution D3 far from D1 instead of near D2 (ablation) | PWF · synthetic SCMs | MSE | Hurt |
| evaluate pre-trained PFNs on uncovered instead of covered interventions | pre-trained PFNs · Lalonde | MSE: error > 5 | Hurt |
| apply meta-sampling fine-tuning (MSF) | pre-trained PFNs · Interaction SCM, |X|=10 | MAE | Improved |
| increase MSF sampling budget K from 5 to 11 (ablation) | MSF · synthetic SCMs | MAE | Improved |
Closest papers by meaning
- Do-PFN: In-Context Learning for Causal Effect Estimation (NeurIPS 2025 · similarity 0.91)
- CausalPFN: Amortized Causal Effect Estimation via In-Context Learning (NeurIPS 2025 · similarity 0.91)
- Estimating Interventional Distributions with Uncertain Causal Graphs through Meta-Learning (NeurIPS 2025 · similarity 0.89)
- Learning Mixtures of Unknown Causal Interventions (NeurIPS 2024 · similarity 0.89)
- Learning Structural Causal Models from Ordering: Identifiable Flow Models (AAAI 2025 · similarity 0.88)
- Generalized Bayes for Causal Inference (NeurIPS 2026 · similarity 0.88)
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.
