Transcript Information Model

Distributed Models for
Decision Support
Jose Cuena &
Sascha Ossowski
Pesented by:
Gal Moshitch &
Rica Gonen
Motivation

The outside world is full of systems which are
governed by complex laws of behavior.

Those systems can be:
– Unanimated entities governed by laws of physics
– Organizations of humans with artificial process
rules.

Often there is a need to influence their dynamics into a
desired direction.
Motivation


For example:
– Computer networks - managed in order to maintain
upper bound on message delays.
–
Road traffic flows - influenced to avoid traffic jams.
–
Air traffic control – influence the planes’ routes to
avoid accidents.
The goal:
– Maximize efficiency.
– Minimize negative impact of faults.
Motivation
Increasing data volume
Decreasing time horizon
Computer applications
support the responsible person
=
Decision support systems (DSS)
Outline
We will discuss:
1. Construction principles of DSS.
2.
Distributed AI (DAI) models and architectures
that applied in DSS.
3.
Applications for energy management and
traffic management.
Construction Principles of DSS.

Modeling DSS:
– A set of world states S
 given by values of the state and control
variables.

– S ideal states,
S  undesired states,
S,S  S
 State
values and control variables that should be
achieved or avoided.
Construction Principles of DSS.

Modeling DSS: (cont.)
– A notion of preference
on states
 “How close” one state is to another - Partial
order/Metric
–
A set  of control actions
 Control variables are changed directly
 State variables are modified indirectly during the
system evolution
Construction Principles of DSS.

Crucial questions DSS should know the answer on:
– What is happening?
 “understand” a situation by identifying
advantageous and problematic aspects.
– What may happen?
 The evolution of the system if no intervention
takes place.
– What should be done?
 Which are the most convenient actions improve
the results.
Knowledge-Based DSS

KnowledgeBased DSS
apply - divide
and conquer
strategy.

An example of a
task-methodssubtasks tree
(TMST).
Knowledge-Based DSS

Task-oriented modeling:
–
The classification task


–
The diagnosis task

–
classifies the situation with respect to its desirability.
Output set of problematic features of the current situation.
An explanation that identifies the causes of such
undesirable behavior.
The prediction task

Evaluates how state S will evolve into state S’ given certain
values for the control variables.
Knowledge-Based DSS

Task-oriented modeling:(cont.)
–
The option generation task

–
Generates a set of plans to overcome the problems
identified.
The action selection task

Selects which of the potential plans will be the outcome of
the management process.
Distributed AI (DAI) Models


Agent-based structuring introduces a more complex
notion of modularity to computer science.
Notion of agents allows:
– Level of specialty

–
Level of autonomy


Designing agents that specialized in basic functions
Integrate in an agent a set of functions required for the
whole application but limited in scope. (i.e. time, space).
Generality of agent allows:
–
Human principles for structuring organizations as design
criteria.
Distributed AI (DAI) Models



The coordination problem has two solutions approach:
Centralized
– Special coordinator agent responsible for detecting
interdependencies between the local agents’
activities.
Decentralized
– No such special agent exists
– Agents interact laterally
– They have the knowledge to discover
inconsistencies between their intended actions and
mutually adapt their local decisions.
Distributed AI (DAI) Models
Distributed AI (DAI) Models

Centralized approach:
–
All possible cases of inconsistencies analyzed a
priori and taken into account by upper level
modules.
–
Disadvantage
 If additional lower level models are introduced, a
sequence of changes has to be produced in the
upper level models.
Distributed AI (DAI) Models

Decentralized approach:
–
Advantages:




–
Systems that are easier to build
(defined very accurately only at the local level)
Easy maintenance
Stable coexistence independent of the number of agents in
society.
No problems of propagation to upper levels appear.
Disadvantage:

Quality of the intelligence of the whole society of agents.
Distributed AI (DAI) Architectures
for DSS.

The architecture does not consider computation and
efficiency.

It considers only features necessary for different case
studies.
Distributed AI (DAI) Architectures
for DSS.

The architecture is built around three major
components:
– A perception subsystem

–
An intelligence subsystem

–
Allows the agent to be situated in the environment and in
society by perceiving agent messages.
Manages the different aspects of information processing as
well as individual and social problem-solving.
An action subsystem



Enacts the plans produced by the intelligent subsystem
Displaying messages to the control personal
Sending messages to other agents
Distributed AI (DAI) Architectures
for DSS.

The architecture is composed of three models:
– Information Model
– Knowledge Model
– Control Model

Information model and knowledge model focuses on
the intelligence subsystem

Control model focuses on the action subsystem.
Information Model

The agents’ dynamic beliefs about the world itself and
the others are stored in the information model.

The perception subsystem writes data on the
information model.

When the intelligence subsystem’s knowledge is
enacted, the information model is modified.

The action subsystem reads from the information
model.
Information Model

The information model composed of two types of
information:
– Problem-solving information
 Local problem-solving tasks information
 Social problem-solving task information
– Control information
 An agenda of what is “intended to be done”
–
Task agenda – keeps track of the tasks to be achieved
locally.
–
Conversation agenda – keeps track of the social methods
in which it is involved.
Knowledge Model

Agent knowledge can be classified from two
perspectives.
– Problem solving knowledge

–
Strategic knowledge


which actions to take
helps to choose among different options that the
intelligence subsystem is to process next.
Agent knowledge can be classified according to its role
– Individual agent knowledge
– Social knowledge
Knowledge Model

Individual agent knowledge:
–
Motivation knowledge

–
A collection of patterns modeling different classes of events
considered by the agent as relevant in the external world.
Local problem-solving knowledge


Basic methods
– perform elementary functions by specific algorithms or
constraints.
Compound methods
– TMST tree, rules or hard-coded simple algorithm.
Knowledge Model

Individual agent knowledge (cont.):
–
Local strategic knowledge
 Generation of the TMST tree.
 At every level and for every task selects the
method to be used.
Knowledge Model

Social knowledge:
– Acquaintance models


–
Knowledge about other agents is stored in these models.
By application of a pattern matching method it can be
deduced whether and up to which degree some
acquaintance provides desired characteristics.
Social strategic knowledge


Determines the next conversation to work on.
Generation of the TMST tree when methods of different
agents integrated.
Knowledge Model

Social knowledge (cont.):
– Social methods:
 Copes with a task by solving its subtasks
 Specify at a very high level how these subtasks
are to be integrated.
 Task assignment
–
Selection of an agent, when several available.
Knowledge Model

Social knowledge (cont.):
– Social methods (cont.):
 Task synchronization
–
Once tasks are assigned, the flow of information
between them needs to be synchronized.
 Solution
–
integration
The results of subtasks of a social method need to be
adapt to each other in order to receive a consisting
result.
Control Model
Perception Subsystem
Intelligence Subsystem
Strategic Knowledge
Motivation
Knowledge
Local Strategic
Knowledge
Acquaintance
Models
Social Strategic
Knowledge
Action Subsystem
Information Model
Messages
Perceptions
Local Problem Social Problem
Information
Solving
Task Agenda
Conversation
Agenda
Problem Solving Knowledge
Local Problem
Solving
Knowledge
Local
Social Methods
Social
Problem
Solving
Inf
Messages
Control
Inf
Actions
Control Model
Perception Subsystem
Intelligence Subsystem
Strategic Knowledge
Motivation
Knowledge
Local Strategic
Knowledge
Acquaintance
Models
Social Strategic
Knowledge
Action Subsystem
Information Model
Messages
Perceptions
Local Problem Social Problem
Information
Solving
Task Agenda
(Add)
Conversation
Agenda
Problem Solving Knowledge
Local Problem
Solving
Knowledge
Local
Social Methods
Social
Problem
Solving
Inf
Messages
Control
Inf
Actions
Control Model
Perception Subsystem
Intelligence Subsystem
Strategic Knowledge
Motivation
Knowledge
Local Strategic
Knowledge
Acquaintance
Models
Social Strategic
Knowledge
Action Subsystem
Information Model
Messages
Perceptions
Local Problem Social Problem
Information
Solving
Task Agenda
(Add)
Conversation
Agenda
Problem Solving Knowledge
Local Problem
Solving
Knowledge
Local
Social Methods
Social
Problem
Solving
Inf
Messages
Control
Inf
Actions
Control Model
Perception Subsystem
Intelligence Subsystem
Strategic Knowledge
Motivation
Knowledge
Local Strategic
Knowledge
Acquaintance
Models
Social Strategic
Knowledge
Action Subsystem
Information Model
Messages
Perceptions
Local Problem Social Problem
Information
Solving
Task Agenda
(Execute Sum)
Conversation
Agenda
Problem Solving Knowledge
Local Problem
Solving
Knowledge
Local
Social Methods
Social
Problem
Solving
Inf
Messages
Control
Inf
Actions
Control Model
Perception Subsystem
Intelligence Subsystem
Strategic Knowledge
Motivation
Knowledge
Local Strategic
Knowledge
Acquaintance
Models
Social Strategic
Knowledge
Action Subsystem
Information Model
Messages
Perceptions
Local Problem Social Problem
Information
Solving
Task Agenda
Conversation
Agenda
Problem Solving Knowledge
Local Problem
Solving
Knowledge
Local
Social Methods
Social
Problem
Solving
Inf
Messages
Control
Inf
Actions
Control Model
Perception Subsystem
Intelligence Subsystem
Strategic Knowledge
Motivation
Knowledge
Local Strategic
Knowledge
Acquaintance
Models
Social Strategic
Knowledge
Action Subsystem
Information Model
Messages
Perceptions
Local Problem Social Problem
Information
Solving
Task Agenda
Conversation
Agenda
Problem Solving Knowledge
Local Problem
Solving
Knowledge
Local
Social Methods
Social
Problem
Solving
Inf
Messages
Control
Inf
Actions