Graphs for margins of Bayesian networks

Abstract

Directed acyclic graph (DAG) models, also called Bayesian networks, impose conditional independence constraints on a multivariate probability distribution, and are widely used in probabilistic reasoning, machine learning and causal inference. If latent variables are included in such a model, then the set of possible marginal distributions over the remaining (observed) variables is generally complex, and not represented by any DAG. Larger classes of mixed graphical models, which use multiple edge types, have been introduced to overcome this; however, these classes do not represent all the models which can arise as margins of DAGs. In this paper we show that this is because ordinary mixed graphs are fundamentally insufficiently rich to capture the variety of marginal models. We introduce a new class of hyper-graphs, called mDAGs, and a latent projection operation to obtain an mDAG from the margin of a DAG. We show that each distinct marginal of a DAG model is represented by at least one mDAG, and provide graphical results towards characterizing when two such marginal models are the same. Finally we show that mDAGs correctly capture the marginal structure of causally-interpreted DAGs under interventions on the observed variables.

References in corpus (14)

• Beyond Bell's Theorem: Correlation Scenarios
• Likelihood ratio tests and singularities
• On the Testability of Causal Models with Latent and Instrumental Variables
• Statistics, Causality and Bell's Theorem
• Graphs for margins of Bayesian networks
• Instrumentality Tests Revisited
• Markovian acyclic directed mixed graphs for discrete data
• Margins of discrete Bayesian networks
• A Constraint Propagation Approach to Probabilistic Reasoning
• Probability distributions with summary graph structure
• On the Testable Implications of Causal Models with Hidden Variables
• Inferring latent structures via information inequalities
• Sparse Nested Markov models with Log-linear Parameters
• Maximum likelihood fitting of acyclic directed mixed graphs to binary data

Cited by in corpus (34)

• Foundations of Structural Causal Models with Cycles and Latent Variables
• Causal Compatibility Inequalities Admitting Quantum Violations in the Triangle Structure
• Graphs for margins of Bayesian networks
• Nested Markov Properties for Acyclic Directed Mixed Graphs
• Margins of discrete Bayesian networks
• The Inflation Technique Completely Solves the Causal Compatibility Problem
• Markov Properties for Graphical Models with Cycles and Latent Variables
• Causal Structure Learning: a Combinatorial Perspective
• What-is and How-to for Fairness in Machine Learning: A Survey, Reflection, and Perspective
• Markov equivalence of marginalized local independence graphs
• Experimental nonclassicality in a causal network without assuming freedom of choice
• Constraint-based Causal Discovery for Non-Linear Structural Causal Models with Cycles and Latent Confounders
• Which causal structures might support a quantum-classical gap?
• Semidefinite tests for latent causal structures
• Learning a Structural Causal Model for Intuition Reasoning in Conversation
• A convergent inflation hierarchy for quantum causal structures
• Causal Calculus in the Presence of Cycles, Latent Confounders and Selection Bias
• Efficient adjustment sets for population average treatment effect estimation in non-parametric causal graphical models
• Witnessing Non-Classicality in a Simple Causal Structure with Three Observable Variables
• Causal Discovery with Unobserved Confounding and non-Gaussian Data
• A Combinatorial Solution to Causal Compatibility
• Classifying Causal Structures: Ascertaining when Classical Correlations are Constrained by Inequalities
• Causal reappraisal of the quantum three box paradox
• On the Properties of MVR Chain Graphs
• Learning latent causal graphs via mixture oracles
• Causal Networks and Freedom of Choice in Bell's Theorem
• Latent-free equivalent mDAGs
• Distinguishing quantum causal scenarios with indistinguishable classical analogs: The significance of intermediate latents
• Partial Identifiability in Discrete Data With Measurement Error
• Quantum non-classicality in the simplest causal network
• Variable elimination, graph reduction and efficient g-formula
• Probability Based Independence Sampler for Bayesian Quantitative Learning in Graphical Log-Linear Marginal Models
• Fully quantum inflation: quantum marginal problem constraints in the service of causal inference
• Towards Characterising Bayesian Network Models under Selection