Knowledge-Based Decision Model Construction for Dynamic Interpretation Tasks.

摘要

The aim of this project was to identify general principles and develop concrete techniques for knowledge-based construction of probabilistic models supporting dynamics decision making under uncertainty. We focused on problems where the precis decision context (i.e., which options are available and what information is known) is highly variable, precluding specification of a fixed model in advance. The project yielded technical results in four areas of reasoning and decision making under uncertainty involving model construction: (1) path planning and scheduling under uncertainty, (2) abstraction and other approximation methods for Bayesian networks, (3) Bayesian methods for pattern and plan recognition, and (4) aggregation of beliefs across multiple agents.