Lecture 10: Knowledge Representation

Lecture 10: Knowledge Representation¶

AIMA Chapter 10 — 1 hour¶

Learning Objectives¶

• Understand ontological engineering

• Represent categories, objects, and events

• Use semantic networks and description logics

• Handle default and uncertain information

Ontological Engineering¶

• Ontology: Specification of conceptualization

• Categories: Classes of objects

• Generalization: Hierarchy (is-a, part-of)

Categories and Objects¶

• Physical composition: Parts and wholes

• Measurements: Height, weight, units

• Things vs. stuff: Count nouns vs. mass nouns (chair vs. water)

Events¶

• Events: Occurrences with time and participants

• Fluents: Properties that hold over intervals

• Temporal reasoning: Before, during, after

Time¶

• Points: Instants

• Intervals: Durations

• Allen’s algebra: 13 relations between intervals

Mental Objects¶

• Beliefs, desires, intentions: BDI

• Modal logic: □ (necessarily), ◇ (possibly)

• Epistemic logic: Kₐφ (agent a knows φ)

Semantic Networks¶

• Nodes: Objects, categories

• Links: is-a, part-of, has-property

• Inheritance: Properties propagate down hierarchy

• Multiple inheritance: Conflicts

Description Logics¶

• Formal: Decidable subset of FOL

• Concepts: Unary predicates

• Roles: Binary relations

• Reasoning: Subsumption, instance checking

Default Reasoning¶

• Default: “Birds fly” (unless penguin)

• Nonmonotonic: New info can invalidate conclusions

• Circumscription: Minimize exceptions

• Default logic: Normal vs. abnormal

Truth Maintenance Systems¶

• Belief revision: Update when new info

• Justification: Why we believe

• Dependency-directed backtracking: Retract minimal set

Summary¶

• Ontology: Categories, objects, events

• Semantic networks: Inheritance

• Description logics: Decidable reasoning

• Defaults: Nonmonotonic

References¶

• Russell & Norvig, AIMA 4e, Ch. 10

• Chapter PDF: chapters/chapter-10.pdf

Questions?¶

Next lecture: Automated Planning (Chapter 11)