Publication:
Method for Enabling Causal Inference in Relational Domains

Simple item page
dc.contributor.advisorDavid Jensen
dc.contributor.advisorBen Marlin
dc.contributor.advisorDan Sheldon
dc.contributor.advisorNick Reich
dc.contributor.authorArbour, David
dc.contributor.departmentUniversity of Massachusetts Amherst
dc.date2023-09-23T17:58:11.000
dc.date.accessioned2024-04-26T16:18:55Z
dc.date.available2024-04-26T16:18:55Z
dc.date.submittedMay
dc.date.submitted2017
dc.description.abstractThe analysis of data from complex systems is quickly becoming a fundamental aspect of modern business, government, and science. The field of causal learning is concerned with developing a set of statistical methods that allow practitioners make inferences about unseen interventions. This field has seen significant advances in recent years. However, the vast majority of this work assumes that data instances are independent, whereas many systems are best described in terms of interconnected instances, i.e. relational systems. This discrepancy prevents causal inference techniques from being reliably applied in many real-world settings. In this thesis, I will present three contributions to the field of causal inference that seek to enable the analysis of relational systems. First, I will present theory for consistently testing statistical dependence in relational domains. I then show how the significance of this test can be measured in practice using a novel bootstrap method for structured domains. Second, I show that statistical dependence in relational domains is inherently asymmetric, implying a simple test of causal direction from observational data. This test requires no assumptions on either the marginal distributions of variables or the functional form of dependence. Third, I describe relational causal adjustment, a procedure to identify the effects of arbitrary interventions from observational relational data via an extension of Pearl's backdoor criterion. A series of evaluations on synthetic domains shows the estimates obtained by relational causal adjustment are close to those obtained from explicit experimentation.
dc.description.degreeDoctor of Philosophy (PhD)
dc.description.departmentComputer Science
dc.identifier.doihttps://doi.org/10.7275/10006901.0
dc.identifier.orcidN/A
dc.identifier.urihttps://hdl.handle.net/20.500.14394/20193
dc.relation.urlhttps://scholarworks.umass.edu/cgi/viewcontent.cgi?article=2018&context=dissertations_2&unstamped=1
dc.source.statuspublished
dc.subjectCausal Inference
dc.subjectRelational Learning
dc.subjectMachine Learning
dc.subjectArtificial Intelligence and Robotics
dc.titleMethod for Enabling Causal Inference in Relational Domains
dc.typeopenaccess
dc.typearticle
dc.typedissertation
digcom.contributor.authorisAuthorOfPublication|email:darbour@cs.umass.edu|institution:University of Massachusetts Amherst|Arbour, David
digcom.identifierdissertations_2/926
digcom.identifier.contextkey10006901
digcom.identifier.submissionpathdissertations_2/926
dspace.entity.typePublication
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
Thesis__14_.pdf
Size:
1.77 MB
Format:
Adobe Portable Document Format
Name: Thesis__14_.pdf
Size: 1.77 MB
Kind: Adobe PDF
Collections
Dissertations