Quantifying intrinsic causal contributions via structure preserving interventions
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2020
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909217710931968 |
|---|---|
| author | Janzing, Dominik Blöbaum, Patrick Mastakouri, Atalanti A. Faller, Philipp M. Minorics, Lenon Budhathoki, Kailash |
| author_facet | Janzing, Dominik Blöbaum, Patrick Mastakouri, Atalanti A. Faller, Philipp M. Minorics, Lenon Budhathoki, Kailash |
| contents | We propose a notion of causal influence that describes the `intrinsic' part of the contribution of a node on a target node in a DAG. By recursively writing each node as a function of the upstream noise terms, we separate the intrinsic information added by each node from the one obtained from its ancestors. To interpret the intrinsic information as a {\it causal} contribution, we consider `structure-preserving interventions' that randomize each node in a way that mimics the usual dependence on the parents and does not perturb the observed joint distribution. To get a measure that is invariant with respect to relabelling nodes we use Shapley based symmetrization and show that it reduces in the linear case to simple ANOVA after resolving the target node into noise variables. We describe our contribution analysis for variance and entropy, but contributions for other target metrics can be defined analogously. The code is available in the package gcm of the open source library DoWhy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2007_00714 |
| institution | arXiv |
| publishDate | 2020 |
| record_format | arxiv |
| spellingShingle | Quantifying intrinsic causal contributions via structure preserving interventions Janzing, Dominik Blöbaum, Patrick Mastakouri, Atalanti A. Faller, Philipp M. Minorics, Lenon Budhathoki, Kailash Artificial Intelligence Information Theory Machine Learning We propose a notion of causal influence that describes the `intrinsic' part of the contribution of a node on a target node in a DAG. By recursively writing each node as a function of the upstream noise terms, we separate the intrinsic information added by each node from the one obtained from its ancestors. To interpret the intrinsic information as a {\it causal} contribution, we consider `structure-preserving interventions' that randomize each node in a way that mimics the usual dependence on the parents and does not perturb the observed joint distribution. To get a measure that is invariant with respect to relabelling nodes we use Shapley based symmetrization and show that it reduces in the linear case to simple ANOVA after resolving the target node into noise variables. We describe our contribution analysis for variance and entropy, but contributions for other target metrics can be defined analogously. The code is available in the package gcm of the open source library DoWhy. |
| title | Quantifying intrinsic causal contributions via structure preserving interventions |
| topic | Artificial Intelligence Information Theory Machine Learning |
| url | https://arxiv.org/abs/2007.00714 |