Modeling Causal Mechanisms with Diffusion Models for Interventional and Counterfactual Queries
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2023
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913539530162176 |
|---|---|
| author | Chao, Patrick Blöbaum, Patrick Patel, Sapan Kasiviswanathan, Shiva Prasad |
| author_facet | Chao, Patrick Blöbaum, Patrick Patel, Sapan Kasiviswanathan, Shiva Prasad |
| contents | We consider the problem of answering observational, interventional, and counterfactual queries in a causally sufficient setting where only observational data and the causal graph are available. Utilizing the recent developments in diffusion models, we introduce diffusion-based causal models (DCM) to learn causal mechanisms, that generate unique latent encodings. These encodings enable us to directly sample under interventions and perform abduction for counterfactuals. Diffusion models are a natural fit here, since they can encode each node to a latent representation that acts as a proxy for exogenous noise. Our empirical evaluations demonstrate significant improvements over existing state-of-the-art methods for answering causal queries. Furthermore, we provide theoretical results that offer a methodology for analyzing counterfactual estimation in general encoder-decoder models, which could be useful in settings beyond our proposed approach. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2302_00860 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Modeling Causal Mechanisms with Diffusion Models for Interventional and Counterfactual Queries Chao, Patrick Blöbaum, Patrick Patel, Sapan Kasiviswanathan, Shiva Prasad Machine Learning Methodology We consider the problem of answering observational, interventional, and counterfactual queries in a causally sufficient setting where only observational data and the causal graph are available. Utilizing the recent developments in diffusion models, we introduce diffusion-based causal models (DCM) to learn causal mechanisms, that generate unique latent encodings. These encodings enable us to directly sample under interventions and perform abduction for counterfactuals. Diffusion models are a natural fit here, since they can encode each node to a latent representation that acts as a proxy for exogenous noise. Our empirical evaluations demonstrate significant improvements over existing state-of-the-art methods for answering causal queries. Furthermore, we provide theoretical results that offer a methodology for analyzing counterfactual estimation in general encoder-decoder models, which could be useful in settings beyond our proposed approach. |
| title | Modeling Causal Mechanisms with Diffusion Models for Interventional and Counterfactual Queries |
| topic | Machine Learning Methodology |
| url | https://arxiv.org/abs/2302.00860 |