Transcript Presentazione di PowerPoint

HID, CAMO Seminars Series
Top-Down Incremental Development of
Agents' Architecture for Emergency
Management Systems:
TOGA methodology
Andrea Caputo, Adam Maria Gadomski, Franco Delli Priscoli
May 2005
University of Rome “La Sapienza”
Italian National Research Agency ENEA
This activity is realized in cooperation between La Sapienza University and ENEA: F.Delli Priscoli (Univ. La
Sapienza, Rome), A.M.Gadomski (CAMO, ENEA), A.Caputo - thesis (Univ. La Sapienza - Engineering Dep.,
ENEA scholarship 2002/0362)
Top-Down Incremental Development of Intelligent
Agents' Architecture
Presentation outline
• Intelligent Agents' Architecture: Problem Specification
• Existing Design & Programming styles (short soa)
• TOGA Theoretical Tool
• Method: Top-Down incremental development
• Emergency Management Test-Case
• Conclusions
• Prototype demonstration
Contents of the Caputo’s Thesis
• General request overiview
• Contest of the simulation: Socio-Cognitive Engineering
• A TOGA proposal
• IPK monad
• Universal Management Paradigms
• Example showed at SCEF 2003
• Intelligent Decision Support System
• Modelling Disaster Domain
• Disaster Propagation
• GEA
Contest of the Simulation
Socio-Cognitive
Engineering
Natural
Artificial
Sciences
Intelligence
Software
Technology
From the Socio-cognitive contest we will arrive at a
ripetitive, incremental, ricorsive, distribuite
INTELLIGENT ENTITY [ 1 ]
SOCIO-COGNITIVE ENGINEERING PARADIGMS
A TOGA PROPOSAL [ 2 ]
IPK
 Informations
 Preferences
 Knowledges
I’ = Kx I
Kx  K
Kx = Ps (K, I)
( I)
(P)
(K)
I, I’  DD
I
P
K
UMP
Universal Management Paradigm
(UMP) is a functional architecture
of organizational High-Intelligence
for every natural and artificial HighIntelligent agents’ organization.
It is characterized by:

Complete

Relative

Recursive

Incremental
IPK paradigm and UMP describe essential functional properties of
abstract highly intelligent entities, natural and artificial.
TOGA Normative Meta-Assumptions
 structural assumptions:
-- Recursivity
-- Iterativness
-- Repetitivity
-- Modularity
They intend to minimize total axiomatic information employed by the theory.
 methodological assumptions, which require completeness and congruence
of the problem conceptualization on every abstraction level.
 terminological assumption, to reduce the number of terms as is possible.
The key TOGA paradigms (top assumptions/axioms) are divided on [ 3 ] :
Conceptualization, Ontological, and Methodological
TOGA Meta-Modeling Framework
Summarizing, what is it ?
• Complex-Knowledge Ordering Methodology (Meta-theory)
• Problem Specification & Decision-Making Modelling Approach. (It has algebra property)
Three components:
TAO :
Basic conceptualization frame
independent on represented domain of interest.
KNOCS : Axioms system for the real-world problem representation
MRUS :
Methodological RUles Systems
Non ordered
observations,
knowledge,
values
TAO
Conceptualizations
KNOCS
Conceptualization
Goal-oriented
Problem
Model
MRUS: Methodological Rules System
They refers to an Abstract Intelligent Agent (AIA), his/her/its Domain-of-Activity
and to the relations between them.
Personois: IPK Abstract Agent
• Model Axioms
Repetivety
Modularity
I
I LEVEL
Recursivity
…
K
P
II META-LEVEL
P
K
P
K
Universal Management Paradigm
Ref. [ 4 ]
SUPERVISOR
TASKS
INFORMATION
EXPERTISES
COOPERATION
COOPERATING
MANAGER
MANAGER
ADVISOR
TASKS
INFORMATION
INFORMER
Based Structure:
Subjective,
Incremental,
Recursive
EXECUTOR
DISASTER DOMAIN
Disaster Manager: simple model example
Infrastructure Network
Real Emergency Domain
Agent 1
Agent 2
I1
Agent 3
Agent n
I2
I3
In
---
P
K
P
K
P
K
P
K
I : Information
I
P : Preferences
K : Knowledge
Agent Manager
P
K
Objectives of experiment: why?
Practical vefification of the methodology by the
designing a series of agents with incremental
complexity and functionality.
The prototypes have been developed in Object oriented C++
language.
As a test case, we assumed an emergency situation caused by
An explosion in a chemical plant where its consequences cause
An intoxication of the water in a neighboring city.
Definition of the Experiment Architecture
On the base of the TOGA paradigms, we built an evolution line of the
incremental design of Intelligent Agents aimed at the development of the model
of an Intelligent Entity
The representation of the abstract world of the Agent is:
WORLD
ANIMATOR
WORLD
SIMULATOR
PROTOPERSONOID
PERSONOID
ANIMATOR
ABSOLUTE
OBSERVER
In this image is showed the relations between the world of the
Agent and the Human Utent. There are distinghished three
different human roles, evidenced in the lighter boxes
EXPERIMENT: Architecture incrementing
To describe the World Simulator and the Proto-Personoid and the
interaction between them, will be used the following symbolization
DOMAIN
SUPERVISOR
COOPERATING
ADVISOR
Constrain
INFORMER
Environment Domain Body
MANAGER
I
PEXECUTOR
K
World
Personoid
Animator
Animator
Absolute
Observer
Decomposition of different fields of the Agent
The IPK structure is seen from the social prespective according
to the UMP paradigm. Infact in the Domain we can see the
other different components of the UMP paradigm.
IDSS: Intelligent Decision Support Systems
IDSS:
“Software program that integrates human intellectual and computer capacities
to improve decision making quality, in semi-structured problems situations”
[Keen, Scott-Morton, 1996]
DSS
Provides passive Informational Aid and Toolkits
IDSS
Provides active, partially autonomous Decisional Aid
which involve human-like computational intelligence.
When IDSS is important?
• amount of information necessary for the management is so large, or its time
density is so high, that the probability of human errors under time constrains is not
negligible.
• coping with unexpected situation requires remembering, mental elaboration and
immediate application of complex professional knowledge, which if not properly used,
causes fault decisions.
Modelling Disaster Domain: Disaster Prop. Map
Experiment Realization
We created a general agent, which follows a
simple set of rules. It represents a first interaction
of the proto-personoid with the external world.
Then, from this generic starting point, we
decompose the various aspects of the agent,
analysing the IPK monad which represent the core
of the agent. The monad, as we said, is composed
of three different parts (Information, Preferences
and Knowledge), and in every new step of our
decomposition, we increase the complexity of one
of these parts.
To focus this aspect of the analysis we introduce a
scale of colours which represent the grade of the
complexity of the analysed part of the system.
0
1
2
3
4
5
RESULT S OF THE EXPERIMENT
Proto-Personoids produced in the design experiment
The main important results of the experiment are:
 modular and reproducible decomposition of the Personoid has been realized.
 it’s possible to obtain incrementally new specializations of the Personoid focalized on
a more detailed problems
 The complexity of the problem ( functionality and architecture) can growth infinitely.
Test Case: Disaster Domain
Application of Emergency/Disaster Propagation
Framework
Events:
Explosion and fire in chemical factory, Fire in the forest
Emision of toxical substances by tubes to the river
Water in City Aqueduct is toxic
Water users are in danger.
EMERGENCY MANAGER:
Identification of intervention/vulnerable objects, goal of
intervention and possible actions
Test Case: Disaster Propagation Map (DPM)
TEST Case: Time Diagram without intervention
PROPAGATION OF EMERGENCY WITHOUT INTERVENTION
Evolution of the DPM without intervention
Forest
Others
Chicken
Farm
Factory
Factory
tubes
Citizens
River
City
Aqueduct
Combined together the DPM with the Time Diagram
without intervention, this evolution in time will be obtained
GEA: IPK Cognitive Agent
Synthesis of the results of the work
• Documentation and validation of the TOGA Theory
• 25 Agents prototype realized
• 30.000 code lines written
• GEA prototype
• User friendly interface
GEA: Demo
Click here for demonstration
References
1.
2.
3. TOGA Meta-theory Web page:
http://erg4146.casaccia.enea.it/wwwerg26701/Gadtoga.htm
4.