Learning to Share Meaning in a Multi-Agent System (Part I)

1
Learning to Share Meaning in a Multi-Agent
System(Part I)

• Ganesh Padmanabhan

2
Article

• Williams, A.B., "Learning to Share Meaning in a
  Multi-Agent System ", Journal of Autonomous
  Agents and Multi-Agent Systems, Vol. 8, No. 2,
  165-193, March 2004. (Most downloaded article in
  Journal)

3
Overview

• Introduction (part I)
• Approach (part I)
• Evaluation (part II)
• Related Work (part II)
• Conclusions and Future Work (part II)
• Discussion

4
Introduction

• One Common Ontology? Does that work?
• If not, what issues do we face when agents have
  similar views of the world but different
  vocabularies?
• Reconciling Diverse Ontologies so that Agents can
  communicate effectively when appropriate.

5
Diverse Ontology Paradgm Questions Addressed

• How do agents determine if they know the same
  semantic concepts?
• How do agents determine if their different
  semantic concepts actually have the same
  meaning?
• How can agents improve their interpretation of
  semantic concepts by recursively learning missing
  discriminating attributes?
• How do these methods affect the group
  performance at a given collective task?

6
Ontologies and Meaning

• Operational Definitions Needed
• Conceptualization, ontology, universe of
  discourse, functional basis set, relational basis
  set, object, class, concept description, meaning,
  object constant, semantic concept, semantic
  object, semantic concept set, distributed
  collective memory

7
Conceptualization

• All objects that an agent presumes to exist and
  their interrelationships with one another.
• Tuple Universe of Discourse, Functional Basis
  Set, Relational Basis Set

8
Ontology

• Specification of a conceptualization
• Mapping of language symbols to an agents
  conceptualization
• Terms used to name objects
• Functions to interpret objects
• Relations in the agents world

9
Object

• Anything we can say something about
• Concrete or Abstract ? classes
• Primitive or Composite
• Fictional or non-fictional

10
UOD and ontology

• The difference between the UOD and the ontology
  is that the UOD are objects that exist but until
  they are placed in an agents ontology, the agent
  does not have a vocabulary to specify objects in
  the UOD.

11
Forming a Conceptualization

• Agents first step at looking at the world.
• Declarative Knowledge
• Declarative Semantics
• Interpretation Function maps an object in a
  conceptualization to language elements

12
Distributed Collective Memory
13
Approach Overview

• Assumptions
• Agents use of supervised inductive learning to
  learn representations for their ontologies.
• Mechanics of discovering similar semantic
  concepts, translation, and interpretation.
• Recursive Semantic Context Rule Learning for
  improved performance.

14
Key Assumptions

• Agents live in a closed world represented by
  distributed collective memory.
• The identity of the objects in this world are
  accessible to all agents and can be known by the
  agents.
• Agents use a knowledge structure that can be
  learned using objects in the distributed
  collective memory.
• The agents do not have any errors in their
  perception of the world even though their
  perceptions may differ.

15
Semantic Concept Learning

• Individual Learning, i.e. learning ones own
  ontology
• Group Learning, i.e. one agent learning that
  another agent knows a particular concept

16
WWW Example Domain

• Web Page specific semantic object
• Groupings of Web Pages semantic concept or
  class
• Analogous to Bookmark organization
• Words and HTML tags are taken to be boolean
  features.
• Web Page represented by boolean vector.
• Concepts ? Concept Vectors ? Learner ? Semantic
  Concept Description (rules)

17
Ontology Learning

• Supervised Inductive Learning
• Output Semantic Concept Descriptions (SCD)
• SCD are rules with a LHS and RHS etc.
• Object instances are discriminated based on
  tokens contained within sometimes resulting in
  a peculiar learned descriptor vocabulary.
• Certainty Value

18
Locating Similar Semantic Concepts

1. Agent queries another agent for a concept by
   showing it examples.
2. Second agent receives examples and uses its own
   conceptualization to determine if it knows the
   concept (K), maybe knows it (M), or doesnt know
   it (D).
3. For cases, K and M, the second agent sends back
   examples of what it thinks is the concept that
   was queried.
4. First agent receives the examples, and interprets
   those using its own conceptualization to verify
   that they are talking about the same concept.
5. If verified, the querying agent then adds that
   the other agent knows its concept to its own
   knowledge base.

19
Concept Similarity Estimation

• Assuming two agents know a particular concept, it
  is feasible and probable given a large DCM, that
  the sets of concept defining objects differ
  completely.
• Cannot simply assume that the target functions
  generated by each agent using supervised
  inductive learning from example will be the same.
• Need to define other ways to estimate similarity.

20
Concept Similarity Estimation Function

• Input sample set of objects representing a
  concept in another agent
• Output Knows Concept (K), Might Know Concept
  (M), Dont Know Concept(D).
• Set of Objects ? Tries mapping set of objects to
  each of its concepts using description rules ?
  each concept receives an interpretation value ?
  interpretation value is compared with thresholds
  to make K,M, or D determination.
• Interpretation Value for one concept is the
  proportion of objects in the CBQ that were
  inferred to be this concept.
• Positive Interpretation Threshold how often
  this concept description correctly determined an
  object in the training set to belong to this
  concept
• Negative Interpretation Threshold

21

• Group Knowledge
• Individual Knowledge
• Verification

22
Translating Semantic Concepts

• Same algorithm as for locating similar concepts
  in other agents.
• Two concepts determined to be the same, can be
  translated regardless of label in the ontologies.
• Difference After verification, knowledge is
  stored as Agent B knows my semantic concept X as
  Y.