Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Marek J. Druzdzel

NAME:
- Marek J. Druzdzel

DEGREE:
- Ph.D.

DISCIPLINE:
- Engineering Systems and Policy

SCHOOL:
- Carnegie Mellon University (USA) (1993)

ADVISORS:
- None

COMMITTEE MEMBERS:
- None

MPACT Status: Incomplete - Not_Inspected

Title: Probabilistic reasoning in decision support systems: From computation to common sense

Abstract: One of the obstacles in dissemination of normative methods in computer aided decision making is the apparent counterintuitiveness of probability theory. The thesis argues that probability theory is, as Laplace once wrote, "only common sense reduced to computation." The leading theme of the thesis is demonstration that it is possible to retrieve this common sense back from probabilistic reasoning to form foundations for human-computer interfaces. It argues on theoretical and empirical grounds, that it is essential to understand and to explore the interaction between probability and causality. It shows that the mechanism-based view of causality, proposed by Herbert Simon, is applicable to probabilistic domain models. It formulates the conditions under which the structure of a Bayesian belief network can be given a causal interpretation. It also shows the applicability of the mechanism-based view to the work on causal discovery from data. The thesis proposes two views of probabilistic belief updating: belief propagation and scenario thinking. Belief propagation is based on tracing the impact of evidence on the variable of interest through chains of local probabilistic influences. Scenario thinking consists of weighting various deterministic scenarios that support or disprove the hypothesis of interest. The thesis demonstrates that each of these views provides a good basis for automatic generation of explanations in decision support systems. It argues that probability theory rests on sound qualitative foundations that allow for development of normative methods for qualitative reasoning under uncertainty. It proposes a computationally efficient algorithm for qualitative belief propagation in multiply-connected belief networks. Finally, it proposes a scenario-based algorithm for decision-theoretic inference, extends scenario-based explanations to decision making, and argues against the adequacy of the current logic-based AI approaches to reasoning under uncertainty.

MPACT Scores for Marek J. Druzdzel

A = 1
C = 5
A+C = 6
T = 1
G = 1
W = 1
TD = 1
TA = 0
calculated 2012-07-31 12:58:58

Advisors and Advisees Graph