← 研踪 PaperTrailarXiv:2606.17516 · v2
FoundCause: A Foundation Model for Causal Discovery with Latent Confounders from Observational Data
首次提交 2026-06-16 12:50
版本更新 2026-11-17 08:07
雷达观测 2026-11-18 19:43
摘要
Causal discovery from observational data requires recovering directed relations and latent confounding without interventions. We propose FoundCause, an {\em amortized causal discovery foundation model} trained only on synthetic structural causal models that maps datasets to causal graphs in a single forward pass. A permutation-equivariant transformer alternates attention over samples and variables to model cross-variable dependence and per-variable distributions. Pairwise dependence and asymmetry statistics guide its attention, edge decoder and confounder module. The decoder separates edge existence from direction, while triangular refinement incorporates evidence from chains and colliders. Learnable latent tokens predict which variable pairs share a hidden common cause, and a masked input representation handles missing data. To our knowledge, FoundCause is the first amortized causal discovery approach for i.i.d. observational data to predict latent confounding as a separate, explicit output. On 14 real-world and semi-synthetic datasets, FoundCause leads 11 classical and 7 amortized baselines in average AUROC, , skeleton , and normalized structural Hamming distance (): clearly in ( vs.\ ), narrowly in ( vs.\ ). It also has the highest accuracy among amortized methods on bivariate cause--effect pairs.
作者备注:Download the model at https://github.com/amazon-science/foundcause
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-17 08:07 · 最新观测版本
FoundCause: A Foundation Model for Causal Discovery with Latent Confounders from Observational Data
Causal discovery from observational data requires recovering directed relations and latent confounding without interventions. We propose FoundCause, an {\em amortized causal discovery foundation model} trained only on synthetic structural causal models that maps datasets to causal graphs in a single forward pass. A permutation-equivariant transformer alternates attention over samples and variables to model cross-variable dependence and per-variable distributions. Pairwise dependence and asymmetry statistics guide its attention, edge decoder and confounder module. The decoder separates edge existence from direction, while triangular refinement incorporates evidence from chains and colliders. Learnable latent tokens predict which variable pairs share a hidden common cause, and a masked input representation handles missing data. To our knowledge, FoundCause is the first amortized causal discovery approach for i.i.d. observational data to predict latent confounding as a separate, explicit output. On 14 real-world and semi-synthetic datasets, FoundCause leads 11 classical and 7 amortized baselines in average AUROC, , skeleton , and normalized structural Hamming distance (): clearly in ( vs.\ ), narrowly in ( vs.\ ). It also has the highest accuracy among amortized methods on bivariate cause--effect pairs.
查看此版本解读 →查看此版本原文 ↗