Context-Specific Independence in Bayesian Networks
arXiv:1302.3562
Abstract
Bayesian networks provide a language for qualitatively representing the conditional independence properties of a distribution. This allows a natural and compact representation of the distribution, eases knowledge acquisition, and supports effective inference algorithms. It is well-known, however, that there are certain independencies that we cannot capture qualitatively within the Bayesian network structure: independencies that hold only in certain contexts, i.e., given a specific assignment of values to certain variables. In this paper, we propose a formal notion of context-specific independence (CSI), based on regularities in the conditional probability tables (CPTs) at a node. We present a technique, analogous to (and based on) d-separation, for determining when such independence holds in a given network. We then focus on a particular qualitative representation scheme - tree-structured CPTs - for capturing CSI. We suggest ways in which this representation can be used to support effective inference algorithms. In particular, we present a structural decomposition of the resulting network which can improve the performance of clustering algorithms, and an alternative algorithm based on cutset conditioning.
Appears in Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence (UAI1996)
References in corpus (1)
Cited by in corpus (34)
- Decision-Theoretic Planning: Structural Assumptions and Computational Leverage
- The Bayesian Structural EM Algorithm
- SPUDD: Stochastic Planning using Decision Diagrams
- Network Fragments: Representing Knowledge for Constructing Probabilistic Models
- Rao-Blackwellised Particle Filtering for Dynamic Bayesian Networks
- SPOOK: A System for Probabilistic Object-Oriented Knowledge Representation
- Learning Arithmetic Circuits
- Utilities as Random Variables: Density Estimation and Structure Discovery
- Value Elimination: Bayesian Inference via Backtracking Search
- Minimal sufficient causation and directed acyclic graphs
- Loglinear models for first-order probabilistic reasoning
- Computational Advantages of Relevance Reasoning in Bayesian Belief Networks
- Formula-Based Probabilistic Inference
- Contextual Weak Independence in Bayesian Networks
- Context-Specific Approximation in Probabilistic Inference
- Propagation using Chain Event Graphs
- Correlated Action Effects in Decision Theoretic Regression
- Bayesian Error-Bars for Belief Net Inference
- Sufficiency, Separability and Temporal Probabilistic Models
- Toward General Analysis of Recursive Probability Models
- Learning Bayesian Networks with Restricted Causal Interactions
- Structured Arc Reversal and Simulation of Dynamic Probabilistic Networks
- A Separation Theorem for Chain Event Graphs
- Representing and Combining Partially Specified CPTs
- Temporal Action-Graph Games: A New Representation for Dynamic Games
- Approximation by Quantization
- LogitBoost autoregressive networks
- Efficient Inference in Large Discrete Domains
- Inference for Multiplicative Models
- The causal manipulation of chain event graphs
- Approximate Causal Abstraction
- Learning Bayesian Networks with Local Structure
- Using Bayesian Network Representations for Effective Sampling from Generative Network Models
- Likelihood Computations Using Value Abstractions