From Probabilistic Argumentation to Information Algebras (and back)

1 / 17

# From Probabilistic Argumentation to Information Algebras (and back) - PowerPoint PPT Presentation

From Probabilistic Argumentation to Information Algebras (and back). Rolf Haenni Computer Science Department University of California, Los Angeles. Contents :. 1. Introduction 2. Probabilistic Argumentation 3. Dempster-Shafer Theory 4. Valuation and Information Algebras 5. Conclusion.

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.

## PowerPoint Slideshow about 'From Probabilistic Argumentation to Information Algebras (and back)' - chance

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

Rolf Haenni

Computer Science DepartmentUniversity of California, Los Angeles

Contents:

1. Introduction

2. Probabilistic Argumentation

3. Dempster-Shafer Theory

4. Valuation and Information Algebras

5. Conclusion

1) elaborate possible answers or alternatives

2) list pros and cons (for each answer or alternative)

3) measure or weigh pros and cons (for each answer or alternative)

Cons

Pros

1. Introduction

Reasoning and deciding under uncertainty is common in everyone’s daily life:

The most popular formal approach is different:

1) elaborate possible answers or alternatives

2) build probabilistic model (usually a Bayesian network)

3) compute posterior probabilities (for each answer or alternative)

4) apply decision theory (maximize expected utility or minimize expected cost)

 this disrespects:

1) the true nature of uncertain reasoning observed in everyone’s daily life

2) the existence of partial or total ignorance

 e.g. knowing that p(head)=0.5 is different than not knowing the probability p(head)

1. Modeling

Uncertainknowledge

Knowledge

R

a®R

Fact:

a® (R ® S)

R ® S

Simple rule:

a®

General rule:

More general:

Ingredients:

- possible states

- risk elements

- interpretations

- unknown circumst.

- uncertain outcomes

- measurement errors

• propositions
• assumptions
• prop. formulas

2. Probabilistic Argumentation

2. Qualitative Analysis

arguments

pro/contra

Hypothesis

hypothesis

 open question about unknownor future world

a) arguments in favor of

 combination of assumptions proving the hypothesis

b) counter-arguments against

 combination of assumptions disproving the hypothesis

Example:

a1®X

a4

a1 a2

arguments

(a2a3) ®Y

hypothesis

knowledgebase

a1a3

(XY) ®Z

Z

a2 a5

a4®Z

counter-arguments

a3 a5

(a5Y) ®Z

Remarks:

1)

4) support and possibility are non-monotone!

5) and means total ignorance

3. Quantitative Analysis

a) define probability distributionover

 e.g. independent probabilities for each assumption

b) compute degree of support:

 conditional probability that at least one argument is true (given no conflicts)

c) compute degree of possibility:

 one minus conditional probability that at least one counter-argument is true (given no conflicts)

Y

H

E

X

Z

H

Y

E

X

J. Bernoulli: “Ars Conjectandi”, 1713

3. Dempster-Shafer Theory

1. Modeling

a) define variables, domains, and frames

with

b) define mass functions

with

 knowledge base:

2. Quantitative Analysis:

 belief:

 plausibility:

and

Degree ofSupport

Arguments

Belief

Probabilistic

ArgumentationSystem

Dempster-Shafer MassFunctions

Hypothesis

Hypothesis

Degree ofPossibility

Counter-Arguments

Plausibility

Remark:

 Every probabilistic argumentation system can be transformed into a set of mass function such that

for all .

a1

a1®X

X

(a2a3) ®Y

(XY) ®Z

a2

a4

a4®Z

Y

Z

a3

(a5Y) ®Z

a5

4. Valuation and Information Algebras

• Knowledge is usually composed of small pieces of information
• Each piece of information affects only a few variables

set of variables

• valuation  “piece of information”

Framework:

• domain
•  set of valuations with
•  all valuations

1988: Local Computation (for probabilistic inference), Lauritzen & Spiegelhalter, J. Royal Statistical Society

1990: Valuations, Axioms, Propagation in Join TreesShenoy & Shafer, UAI’90 & Readings in Uncertain Reas.

2000: Valuation & Information Algebras, Shenoy & Kohlas, Handbook of Defeasible Reas. & Management of Uncert.

2002: Ordered Valuation Algebras, Resource-BoundedApproximation, Haenni, AI Journal (?)

Knowledge base:

Operations:

• combination:

 (aggregated information)

• marginalization:

 (information focussed to )

Problem of Inference:

• compute

 marginalize the joint valuation to the domain of interest

Valuation algebra:

 system satisfying (A1) to (A6)

• probability potentials (CPT’s from Bayesian network)
• belief potentials (in the sense of Dempster-Shafer)
• possibility functions

Information algebra:

 system satisfying (A1) to (A7)

• propositional logic
• constraint systems
• systems of linear equations and inequalities
• relational algebra
• probabilistic argumentation systems

5. Conclusion

• Probabilistic argumentation is a natural approach to reasoning under uncertainty.
• From a quantitative point of view, probabilistic argumentation is equivalent to Dempster-Shafer theory.
• The abstract framework of valuation and information algebras is a powerful tool that allows to better under-stand the nature of information and its computational challenges.
• Convenient resource-bounded or anytime algorithms have been developed for probabilistic argumentation, Dempster-Shafer theory, and valuation algebras.