Structured Probabilistic Models: Bayesian Networks and Beyond

History

Outline

Probability distributions

Bayesian networks

BN semantics

BN inference (in theory)

BN inference (in practice)

CPCS

Decision Making

Learning

Probabilistic models: applications

Uncertainty is unavoidable

Build on Strength

Model-based reasoning

Exploit domain structure

Scaling up

Problem: Size

Problem: Knowledge engineering

Problem: Inference

Large-scale BNs

Exploit structure!

Beyond Bayesian networks

OOBNs

Inter-object interactions

Probabilistic model

Semantics

Classes

Inheritance

Specific situation models

OOBN inference

Inference: simple approach

Exploit structure

What have we gained…

Abstraction and refinement

Stochastic dynamic system

Dynamic Bayesian networks

Dynamic OOBNs

Dynamic OOBNs: advantages

What about inference?

DBN inference

Unfortunately not…

Revealed structure

Where are we?

Intelligent hospital manager

What do we need?

But we also need …

We need …

OOBNs ? probabilistic frames

Inter-frame dependencies

Structural uncertainty

Number Uncertainty

Reference uncertainty

Structural uncertainty: inference

Summary

Additional pieces

PPT Slide

Recent papers

Email: koller@cs.stanford.edu

Home Page: http://robotics.stanford.edu/~koller