Networked

1
NetworkedIntelligent Software Agents
AgentsNISA

• Ahmed Hambaba
• Professor Computer Engineering
• NISA
• June 7, 2002

2
NISA Mission
MISSION The core of NISA Center is to develop
intelligent software agent and agents
technologies with focuses on integrated Business,
Engineering and Manufacturing. Its Vision is to
serve as a center of excellence for the creation
and dissemination of a systematic body of
knowledge in information engineering systems and
ultimately to impact next-generation products and
service systems. The Center promotes research
program, education, and community services to
contribute to education and research
infrastructure base.
3
SOFTWARE AGENT TECHNOLOGIES
Agent Communication Languages
Support computing technologies
Languages
Traditional and Object Oriented Languages
Agent (scripting) languages
4
INTELLIGENT SOFTWARE AGENT
Learning Adaptation
Agent Communication Languages
Support computing technologies
Languages
Traditional and Object Oriented Languages
Agent (scripting) languages
5
NISA Platform Tool
Applications
Agent communication
Agent management
Content languages
Agent message transport
Abstract Architecture
Communicative Acts
Interaction protocols
6
NISA Objectives

• Activity Areas
• Industrial Research
• Product Solution Development
• Consulting
• Strategic Path
• Horizontal Platforms Tools (NISA)
• Vertical
• Communications Networks
• Diagnosis Prognosis
• Engineering, Manufacturing
• Commerce, Finance

7
NISA Application Types

• User Assistant Applications

• These systems are those that work with, and in
  the interests of, an end- user in order to
  enhance their productivity and to ease the use of
  complex computer-based systems.

• They are differentiated from standard user
  interfaces, in that they are empowered to act at
  least semi-autonomously, and are not merely tools
  that the user uses and controls.
  
  
• User profile learning systems
• Multimodal interface systems
• Personal Digital Assistant or Personal
  Intelligent Communicator applications(e.g.,
  digital telephone secretary)

8
NISA Application Types

• Information Retrieval Applications

• These systems involve all the services needed to
  help the users in finding easily and quickly the
  information they request. This can be achieved
  for example by a society of agents.
• Directory services (yellow and white pages)
• Data Base enquiry
• Information Brokerage
• Media indexing

Service Management Applications

• These are systems that involve configuration and
  delivery of user requested services at the right
  time, cost, and QoS, while observing the required
  security and privacy issues.
• Multimedia services
• Buy/selling services(e.g., information,
  material goods)
• TMN/intelligent network management services
• Trip planning and guidance services(e.g.,
  intermodal route planning, hotel and
  parking-lot reservations, individualised traffic
  guidance, tourism)

9
NISA Application Types

• Business Management Applications

• These systems deal with management of business
  tasks and resources in provision of services and
  carrying out business operations.
• Financial services
• Electronic commerce
• Workflow management
• Office automation
• Computer Supported Cooperative Work
• Telecommuting

10
NISA Application Types
Manufacturing Management Applications

• These systems involve physically embodied agents
  designed to carry out and deal with management of
  tasks and processes in relatively structured
  industrial environments. These processes may
  involve the control of industrial robots and
  machines via software interfaces.
• Industrial Robotics
• Factory automation
• Virtual factory management
• Load Balancing

• Research Applications

• These systems involve using agent technology to
  further research in other (IT) areas.
• Vision processing
• Learning and adaptive systems
• Speech processing
• Distributed knowledge-based systems
• Human-Computer Interface

11
NISA Organization

• The Center shall conduct research, perform
  technology evaluation, provide the
• academic and industrial community with enhanced
  education capability in the field of
• Multi-Agent System Application in Engineering,
  Manufacturing and Software agent
• technology and facilitate information exchange
  and technology transfer.
• The Center will be catalyzed by a small
  investment from SJSU and it is
• primarily supported by Center members, with
  NISA committees taking a supporting role in
  their development and evolution
• Initially five-year program.
• This five-year period allows for the development
  of a strong partnership between the academic
  researches and their industrial and government
  members.
• The NISA center may enter into agreements with
  other universities to participate as additional.

12
NISA Organization
NSF
Director
IndustryAdvisoryBoard
AcademicPolicyCommittee
NISA Committee
Collaborative Institutions
Evaluator
Human Computer Interface (HCI)
Robotics Lab
Software Engineering Lab
MISE Lab Manufacturing Information
Systems Engineering Lab
Client/Server Computing Lab
13
Organizational Structure

• The organizational structure of the Center
  comprises an Industrial Advisory
• Board (IAB), Center Site Director, and a
  University Policy Committee.

The IAB, consisting of one voting member from
each participating company, provides advice on
research priorities and makes recommendations on
projects to be funded.
The Center Directory manages the day-to-day
operation of the Center, acts as a liaison with
member companies as well as the university
administration and, in collaboration with the
IAB, sets goals and future directions of research.
The University Policy Committee, which consists
of senior leaders the university, will assure
that the Centers activities are consistent with
academic policies and procedures of the
Universities.
In addition, the Center has an external NSF
evaluator to monitor and evaluate research
interaction between Center researchers and
company members.
14
Objectives

• The objectives of the NISA Center are to

Conduct research to bring about innovation and
practical solutions by focusing on industrially
relevant research needs
Foster collaborative research projects between
industrial and academic engineers and scientists
Promote interdisciplinary and inter-university
research activities and to nurture students
through testbeds and collaborative projects.
15
Deliverables

• The center will deliver to its industry members
  the following

Development of software/hardware tools for
manufacturing and engineering (e- diagnosis,
e-prognosis, intelligent monitoring) using
web-enabled and wireless technology
Implementation and demonstration of NISA
architecture and its applications assets (machine
manufacturing system, process, etc.)
Test bed projects with collaborating member
companies
Full-time and on-site post-docs and student
resources for specific projects and
Interdisciplinary, well trained and system
oriented engineers.
16
NISA Center Requirements

• Develop a partnership among academe, industry and
  other organizations
• participating in the Center

Consult with Center members to set a defined
research agenda focused on shared research
interests, needs, and opportunities
Have Center members that monitor and advise on
the progress of the research, which speeds
two-way transfer of knowledge between
universities and industry
Have a strong industry/university interaction
program of university, industry, and other
partners that are the primary financial resource
for the Center
Rely on student involvement in high quality
research projects, thus developing students who
are knowledgeable in industrially relevant
research
Membership agreement.
17
NISA Center Requirements
There must be a minimum of six Center members
with a membership fee of 50,000 or higher per
year.
Other membership level categories with lower fees
may be designated to encourage small company
participation in the Center.
An Industrial Advisory Board (IAB) that reviews
ongoing and completed activities and recommends
new projects.
18
Industry Advisory Board (IAB)

• Joe Pinto, Senior VP, Cisco
• Marypat Farrell VP, Rockwell
• Ben Bierman, Sr. Director, Applied Materials
• Bob Togasaki, VP, HP
• Glen Ahmann, Sr. Director, United Defense
• Firth Griffith, Venture Capitalist Beachhead
  Capital
• Mike Shafto, NASA Ames
• BasheerJanjua, CEO, Integnology
• Naoya Kinoshita, President MCC, Japan

19
Industry

• Cisco Systems
• Applied Materials
• United Defense
• NASA Ames Research Center
• MCC Corporation (Japan)
• Integnology
• Rockwell Automation

20
Industry

• Lockheed Martin Space Systems
• Lam Research Lab
• Solectron
• Agilent
• HP
• Intel
• KLA Tencor

21
NISA Technical Committees

• Ahmed Hambaba, Intelligent Software Agents
• Kevin Corker, Human Computer Interface
• M. E. Fayad, Object-Oriented Technology
• Dan Harkey C/S Computing
• Hussein Salama, Content Delivery Networking
• Rod Fatoohi, Computer Networks
• Winncy Du, Robotic Systems
• Moenes Iskarous, Intelligent Software Agents
• Jacob Tsao, Manufacturing Systems

22
University Collaboration
NSF Industry/University Cooperative Research
Center
IMS Center University of Wisconsin at Milwaukee
IMS Center University of Michigan
NISA Center SJSU
23

• GOAL
• NSF - Industry/University Cooperative Research
  Center on Networked-Intelligent Software Agents

24
IMS CENTER _at_ SJSU NISA Group

• Values to the IMS Center
• Bring IT (ISA, Wireless, HMI) and intelligent
  agent focused technologies to IMS Center.
• Adding Semiconductor Industry, Computer, and
  Networking Industries to IMS members
• Foster alliance between the Silicon Valley
  Industry (Applied Materials, Cisco, United
  Defense, NASA) and IMS Industry Members

25
NISA Architecture Overview

• Agents may have a need to obtain a service by
  other entities in the
• system.

There are and in the future there will continue
to be a wealth of non-agent software systems
which provide useful services.
Agents are to be truly useful they must be able
to interface with and control existing software
system such as databases, web- browsers, set-top
boxes, speech synthesis programs and so forth.
Software systems come in all shapes and sizes.
Many different types of interfaces are possible
each with their own particular networking
protocol, strengths and weaknesses.
26
NISA Architecture Overview

• There are a number of emerging distribution
  technologies such as
• CORBA, DCOM and Java-RMI which are creating
  (competing)
• standards for the integration of software systems
  and resources.

27
Layered Model for a Wrapper
Agent
ACL Messages
Wrapper
Mapping to technology
Orb Trader
Java
April
Java
ActiveX
IIOP
RMI
TCP/IP
DCOM
RMI
Web Server
Legacy System
Java Component Server
CORBA Server
Software Services
28
Reference Model of Agent based Adaptation
Agent Platform A
Agent Platform B
Task Agents
Interface Agents
Information Agents
Middle Agents
Task Agents
Interface Agents
Information Agents
Middle Agents
Agent Directory
Message Transport
Service Directory
ACL
Agent Directory
Message Transport
Service Directory
ACL
IIOP
IIOP
WAP
WAP
TCP/IP
TCP/IP
WLAN
WWAN
Wireline
WLAN
WWAN
Wireline
29
General Agent Software Integration Scenario
Broker
query
Wrapper
Client Agent
Software System
invoke
30
Characteristics of Agent
Existing technologies
Competitive XML technologies
Autonomy independency
persistency
proxy/surrogate
Distributed processing Persistent DB/daemon Fixed
purpose proxy
Intelligence Inference
Interaction Dynamic interface
adaptability rationality
Distributed processing Persistent DB/daemon Fixed
purpose proxy
ebXML, WSDL XML Schema, SOAP, UDDI, e-speak
Social ability Cooperation/
collaboration Competition
Dynamic participation
Distributed algorithm Cooperation protocol,
workflow Auction, economic model Service
advertisement
ebXML, e-speak, WSDL CBL, CXML UDDI, e-speak, WSDL
Mobility mobile agent
Mobile object, process
migration
31
Applications
Agent communication
Agent management
Content languages
Agent message transport
Abstract Architecture
Communicative Acts
Interaction protocols
32
Why an Abstract Architecture?
The first purpose of this work is to foster
interoperability and reusability.
Specifically, if two or more systems use
different technologies to achieve some
functional purpose, it is necessary to identify
the common characteristics of the various
approaches. This leads to the identification of
architectural abstractions .
By describing systems abstractly, one can explore
the relationships between fundamental elements of
these agent systems.
From this set of architectural elements and
relations one can derive a broad set of possible
concrete architectures, which will interoperate
because they share a common abstract design.
33
MULTI-AGENT SYSTEM (MAS)
User 1
User 2
User u
Goal and Task Specifications
Results
Interface Agent 1
Interface Agent 2
Interface Agent 3
Solutions
Tasks
Task Agent1
Task Agent2
Task Agent3
Information Integration Conflict Resolution
Replies
Info Service Requests
Middle agent 2
Information Agent 2
Information Agent 1
Advertisements
Answers
Queries
DB1
DB2
DBm
34
MULTI-AGENT SYSTEM (MAS)
Abstract Architecture
Middle Agents
Information Agents
Task Agents
Interface Agents
Message Transport
Agent Directory
Service Directory
ACL
Concrete realization Java
Concrete realization CORBA
Message Transport
Agent Directory
Service Directory
ACL
Message Transport
Agent Directory
Service Directory
ACL
35
MULTI-AGENT SYSTEM (MAS)

• Broker agent An agent which provides the Agent
  Resource Broker
• (ARB) service. There must be at least one such an
  agent in each Agent
• Platform in order to allow the sharing of
  non-agent services.

Agent An Agent is the fundamental actor in a
domain. It combines one or more service
capabilities into a unified and integrated
execution model which can include access to
external software, human users and communication
facilities.
Agent Communication Language A language with
precisely defined syntax, semantics and
pragmatics that is the basis of
communication between independently designed and
developed software agent. ACL is the primary
subject of this part of the FIPA specification.
36
MULTI-AGENT SYSTEM (MAS)
Agent Communication Channel Router The Agent
Communication Channel is an agent which uses
information provided by the Agent Management
System to route messages between agents within
the platform and to agents resident on other
platforms.
Agent Communication System The Agent Management
System is an agent which manages the creation,
deletion, suspension, resumption, authentication
and migration of agents on the agent platform and
provides a white pages directory service for
all agents resident on an agent platform. It
stores the mapping between globally unique agent
names and local transport addresses used by the
platform.
37
MULTI-AGENT SYSTEM (MAS)
Agent Platform An Agent Platform provides an
infrastructure in which agents can be deployed.
An agent must be registered on a platform
in order to interact with other agents on that
platform or indeed other platforms. An AP
consists of three capability sets ACC, AMS and
default Directory Facilitator.
Software System A software entity which is not
conformant to the FIPA Agent Management
specification.
38
MULTI-AGENT SYSTEM (MAS)

• Interface agents interact with users, receive
  user input, and display results.

Task agents help users perform tasks, formulate
problem solving plans and carry out these plans
by coordinating and exchanging information with
other software agents.
Information agents provide intelligent access
to a heterogeneous collection of information
sources.
Middle agents help match agents that request
services with agents that provide services.
39
Background

• NISAs goal in creating agent standards is to
  promote inter-operable
• agent applications and agent systems.

At the heart NISAs model for agent system is
agent communication, where agents can pass
semantically meaningful messages to one another
in order to accomplish the tasks required by the
application.
How these messages are transferred (that is, the
transport)
How those messages are represented (e.g.
s-expressions, bit- efficient binary objects,
XML)
40
Optional attributes of those messages, such as
how to authenticate or encrypt them.
It also became clear that to create agent
systems, which could be deployed in commercial
settings, it was important to understand and to
use existing software environments. These
environments included elements such as
Distributed computing platforms or programming
languages
Messaging platforms
Security services
Directory services, and,
Intermittent connectivity technologies
41
Because the abstract architecture permits the
creation of multiple concrete realizations, it
must provide mechanisms to permit them to
interoperate.
This includes providing transformations for both
transport and encodings, as well as integrating
these elements with the basuc elements of the
environment.
For example, one agent system may transmit ACL
messages using the OMG IIOP protocol. A second
may use IBMs MQ-series enterprise messaging
system. An analysis of these two systems how
senders and receivers are identified, and how
messages are encoded and transferred allows us
to arrive at a series of architectural
abstractions involving messages, encodings,
and addresses.
42
Going from Abstract to Concrete Specifications
Such an architecture cannot be directly
implemented, but instead the forms the basis for
the development of concrete architectural specific
ations.
Such specifications describe in precise detail
how to construct an agent system, including the
agents and the services that they rely upon, in
terms of concrete software artefacts, such
as programming languages, applications
programming interfaces, network protocols,
operating system services, and so forth.
Several realizations have chosen to use this
directory service model.
43
Methodology
This abstract architecture was created by the use
of UML modeling, combined with the notations of
design patterns.
The analysis drew upon many sources

• The abstract notions of agency and the design
  features that flow from this.

• Commercial software engineering principles,
  especially object-oriented techniques, design
  methodologies, development tools and distributed
  computing models.

• Requirements drawn from a variety of applications
  domains.

• Existing NISA specifications and implementations.

• Agent systems and services, including NISA and
  non-NISA designs.

• Commercial important software systems and
  services, such as Java, CORBA, DCOM, LDAP, X.500
  and MQ Series.

44
INTELLIGENT AGENTS
Recently, research concerning the learning and
adaptation of agents in multi-agent systems has
gained considerable attention in both the
Distributed Artificial Intelligence (DAI) and
Machine Learning (ML) communicates.
Agents must often learn in order to dynamically
acquire the knowledge and skills necessary to
improve their individual performance, precision,
efficiency and scope of solvable problems during
run-time.
45
Neural Networks and Agent Learning

• In an agent-based system, agents learn through a
  variety of methods.

• Rote learning immediate and direct implantation
  of knowledge and skills without requiring further
  inferencing or transformation.

• Isolated learning concerned with having agents
  learn by themselves.

46
Modeling Neural Network Knowledge
Although a number of general specifications have
been proposed for modeling the architecture of
trained neural networks, these models lack
a representation that provides meta-knowledge
about a network.
Neural Network Knowledge The primary motivation
for the use of a Neural Network Knowledge is to
facilitate the probability of a wide variety of
neural networks among heterogeneous agents and
environments.
Neural Network Knowledge may be represented in a
number of languages including LISP, XML and KIF
in order to facilitate their portability among
a variety of environments.
47
Communicating Neural Network Knowledge
The identification of a protocol for
communicating neural network knowledge between
agents
A protocol is required that allows a sending
agent to express the context surrounding a
communicated neural network knowledge to a
receiving agent.
48
Communicating Neural Network Knowledge
Source Agent
Target Agent
Neural Network Knowledge
49
Managing Neural Network Knowledge
The specification of a multi-agent architecture
for managing and using neural network knowledge.
This specification defines a core set of
services for creating, storing, managing and
executing neural network knowledge.
50
Placing This Work in Context
The idea is that by using pre-existing knowledge,
we may exploit knowledge reuse in neural networks
that exist in different, but related domains and
potentially save training time on new tasks.
Such transfer has been shown to

• Accelerate learning of the target network

• Reduce the number of required training examples
  of the target network

• Improve the accuracy of the target network

51
What is a Neural Network?

• A neural network is a massive parallel
  distributed processor that has a
• natural propensity for storing experiential
  knowledge and making it
• available for use. It resembles the brain in two
  respects

• Knowledge is acquired by the network through a
  learning process.

• Interneuron connection strengths known as
  synaptic weights are used to store the knowledge.

52
Models of a Neuron
wk1
x1
Wk1
x2
Activation function
Output
Input Signals
uk
?
?(.)
yk
Summing junction
wk1
xp
Synaptic weights
Threshold
53
Learning process
Learning paradigms
Learning algorithms (rules)
Error- correction learning
Bolzmann Learning
Thorndikes law of effect
Hebbian Learning
Competitive Learning
Supervised learning
Reinforcement Learning
Self-organized (unsupervised) learning
54
COMBINING NUMERICAL AND LINGUISTIC INFORMATION
INTO ENGINEERING SYSTEMS AN ADAPTIVE FUZZY SYSTEM
APPROACH

• For most engineering systems, there are two
  important information sources
• sensors which provide numerical measurements of
  variables, and human experts
• who provide linguistic instructions and
  descriptions about the system

Numerical information is represented by numbers
for example 0.25, 1.44, and so on
Linguistic information is represented by words
like small, large, very large, and so forth.
Conventional engineering approaches can only make
use of numerical information and have difficulty
incorporating linguistic information.
Why is linguistic information usually represented
in fuzzy terms?
55
Pure Fuzzy Logic Systems
Fuzzy Rule Base
Fuzzy Interface Engine
fuzzy sets In U
fuzzy sets In V
56
Fuzzy Logic Systems with Fuzzifierand Defuzzifier

• In order to use the pure fuzzy logic system shown
  in the following figure in
• engineering systems where inputs and outputs are
  real valued variables, the most
• straightforward way is to add a fuzzifier to the
  input and a defuzzifier to the output
• of the pure fuzzy logic system.

Fuzzy Rule Base
Fuzzifier
Defuzzifier
y in V
x in U
Fuzzy Rule Base
fuzzy sets in U
fuzzy sets in V
57
SOFT COMPUTING CONSTITUENTS AND CONVENTIONAL
ARTIFICIAL INTELLIGENCE

• Soft computing is an emerging approach to
  computing which parallels are
• remarkable ability of the human mind to reason
  and learn in an
• environment of uncertainty and imprecision.

Methodology Strength
Neural network Learning and adaptation
Fuzzy set theory Knowledge representation via fuzzy if- then rules
Genetic algorithm and simulated annealing Systematic random search
Conventional AI Symbolic manipulation
58
Knowledge Discovery In Databases

• The amount of data being collected in databases
  today far exceeds our ability to reduce and
  analyze data without the use of automated
  analysis techniques.

• Knowledge discovery is defined as the
  non-trivial extraction of implicit, unknown, and
  potentially useful information from data'' . In,
  a clear distinction between data mining and
  knowledge discovery is drawn. Under their
  conventions, the knowledge discovery process
  takes the raw results from data mining (the
  process of extracting trends or patterns from
  data) and carefully and accurately transforms
  them into useful and understandable information.

59
Background

• The following are basic features that all KDD
  techniques share
• All approaches deal with large amounts of data
• Efficiency is required due to volume of data
• Accuracy is an essential element
• All require the use of a high-level language
• All approaches use some form of automated
  learning
• All produce some interesting results

60
KDD Techniques

• Learning algorithms are an integral part of KDD.

• The probabilistic and statistical approaches.

• Classification approaches such as Bayesian
  classification, inductive logic, data
  cleaning/pattern discovery, and decision tree
  analysis.

• Other approaches include deviation and trend
  analysis, genetic algorithms, neural networks,
  and hybrid approaches that combine two or more
  techniques.