DeDiCa: A Scalable Deconfounding Discrete Causal Generative Model
DeDiCa estimates discrete causal effects with hidden confounding using exact, millisecond interventional inference.
DeDiCa is a discrete causal generative framework that parameterizes causal Bayesian networks as tensors and exactly marginalizes hidden confounders. After one maximum-likelihood fit on observational data, it answers identifiable or proxy-identifiable interventional queries. The authors report average-treatment-effect accuracy matching neural causal models, with memory-efficient CPU training and millisecond exact inference, plus better scaling than causal foundation models.
- DeDiCa targets discrete causal inference with hidden confounding.
- It uses a tensorized causal Bayesian network trained once by maximum likelihood.
- ATE estimates match neural causal generative models.
- Inference is exact and reported in milliseconds on CPU.
- It scales better with more observational samples than causal foundation models.
From the original article
1 min read at arxiv.orgWe introduce DeDiCa, a practical framework for efficient causal inference with discrete variables and hidden confounding that revisits classical probabilistic modeling through a tensorized parameterization of causal Bayesian networks, enabling tractable inference.
By Kunchangtai Liang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.