Learning and Reasoning in Hybrid Structured Spaces.

Intro
Title Page
Abstract
Acknowledgments
Contents
Introduction
Motivation
Contributions
Outline of the Thesis
Background
Probabilistic Graphical Models
Bayesian Networks
Markov Networks
Factor graphs
The belief propagation algorithm
Inference by Weighted Model Counting
Propositional satisfiability
Weighted Model Counting
Logical structure
Inference by Weighted Model Integration
Satisfiability Modulo Theories
Weighted Model Integration
Related work
Modelling and inference
Learning
WMI-PA
Predicate Abstraction
Weighted Model Integration, Revisited
Basic case: WMI Without Atomic Propositions
General Case: WMI With Atomic Propositions
Conditional Weight Functions
From WMI to WMIold and vice versa
A Case Study
Modelling a journey with a fixed path
Modelling a journey under a conditional plan
Efficiency of the encodings
Efficient WMI Computation
The Procedure WMI-AllSMT
The Procedure WMI-PA
WMI-PA vs. WMI-AllSMT
Experiments
Synthetic Setting
Strategic Road Network with Fixed Path
Strategic Road Network with Conditional Plans
Discussion
Final remarks
MP-MI
Preliminaries
Computing MI
Hybrid inference via MI
On the inherent hardness of MI
MP-MI: exact MI inference via message passing
Propagation scheme
Amortizing Queries
Complexity of MP-MI
Experiments
Final remarks
lariat
Learning WMI distributions
Learning the support
Learning the weight function
Normalization
Experiments
Final remarks
Conclusion.