Multi-Agent Technologies for Complex Problem Solving

hurtpotatocreekSecurity

Nov 5, 2013 (3 years and 9 months ago)

95 views

Multi
-
Agent Technologies for
Complex Problem Solving

Dr
.
Petr Skobelev


SEC

«
Knowledge Genesis
»

(
Samara
,
Russia
)

Founder and Chairman of Board of Directors

http://www.kg.ru

skobelev@kg.ru


SEC

«
Magenta Technology
»

(
London
,
UK
)

Со
-
Founder

and member of Board of Directors

http://www.magenta
-
technology.com

skobelev@magenta
-
technology.ru



«
Knowledge Genesis
»

Software Engineering Company

Agenda



Short acquaintance with the Samara region and the
company


Why multi
-
agent technologies
?


Examples of successful projects based on multi
-
agent
technologies



Samara Region
е
-
government


Adaptive real
-
time schedulers


Text Understanding


Data mining


Conclusion


Samara, Russia


Region of 3.3 million people


Located on the bank of Volga river


City of 1.2 million people


Second capital of USSR during WWII


National airspace industry centre


High Tech Defence industry centre


Centre of National logistic Network


Railways, air and automobile hub


Educational centre for Volga region


13 Higher Education Institutions


72 Academic Institutions


Traditions and high prestige of engineering
professions


Mature and highly developed IT
-

market


Innovations actively supported by the
Samara Regional Administration

SEC

«
Knowledge Genesis
» (
Russia
)



Established in 1997 in Samara



Originally from
airspace industry and Russian Academy of Sciences



Unique competences in Multi
-
agent systems and Semantic web



Advanced business & technology vision for solving complex
problems



Innovative technologies for distributed decision making support



More than 100 J2EE and .net programmers and engineers



Expertise in large
-
scale systems, web
-
applications, data bases, et
с
.



Affiliated company


Magenta Technology (UK)


2000



Own development platform



International Network of Partners



Strong connections with Universities and Research Institutes



Flexibility and individual approach to each Customer

About the Company


Founded in 2000 together with EU
Investment Funds


Solve complex real
-
world problems using
Multi
-
Agent Systems


Main directions of work:


Systems for scheduling of oil tankers,
trucks, taxis, factories, etc


Internet marketing and advertising


Operational platforms design


Reducing costs & increasing customer
business performance


Clients in USA & UK


Headquartered in London


Development Centre in Samara


Strong connections with Universities in
Russia, UK and USA


Development center in Samara, Russia

Why Multi
-
Agent Technology
?


One of the new critical software technologies



Capable

of applying Fundamental Principles of Self
-
Organization and
Evolution


Provide smart, flexible and pro
-
active software solutions


Based on negotiations, conflicts solving and finding trade
-
offs


“Crack” previously unsolvable problems


Address limitations in existing technology solutions


Allow representing real
-
world objects and processes


Solve problems the way people do

Developments, Products & Technologies



e
-
Government Systems for

Welfare and
Healthcare



Real time GPS
-
based Adaptive

Schedulers



Enterprise Decision Making Support
Systems



Internet Portals



Web
-
based Data Mining

and Text Understanding Systems



Multimedia, 3D
-
Graphics and Animation



Geographic information systems



e
-
Learning Systems

Multi
-
Agent

e
-
Government system

for Social
sphere of the Samara region


Designed for providing targeted
state services based on social
cards of citizens



Multi
-
Agent

e
-
Government system

for Social
sphere of the Samara region





Provides targeted state services



Based on social passports and smart cards of citizens



Knowledge Base contains more than 500 social laws (federal,
regional and municipal): rules applicable to citizen data



Personalized agent attached to each citizen



$YDLODEOH?YLD?WKH?,QWHUQHW?DQG?´,QWHUQHW
-
.LRVNVµ




Knowledge Bases of Social Legislation


Multi
-
agent e
-
government system for social sphere of the Samara
region is based on
Knowledge bases of social legislation

(in the
form of semantic networks) containing
:


Integrated knowledge bases of federal, regional and
municipal laws
;


Regulations for state services provision.



Database

Knowledge

Base

Benefit

Category

Law

Organization

address

Rules

Source of

financing

Human



Name

Year


Post Address

1

Ivanov Ivan
Ivanovich

1934

Samara
,
Sadovaya
Street

34
-
7

Databases vs Knowledge Bases



Databases



Rigid database scheme
,
new
attributes require new
programming


Data organized

as a sequential
indexed arrays


Database elements are data only


Queries are pre
-
defined and
programmed in advance



Effective storage for simple
homogeneous sets of data only

(
for example, years of birth, post
addresses
)



Extensible

«
glossary of terms
»

for
description of new laws and citizen
characteristics



Data represented as a semantic
network


Concepts/relationships and rules can
be included into network


Queries should discover facts and
can be carried out using complex
logical reasoning



Effective storage for diverse data on
citizens (social, medical and other
information)

Knowledge

bases

Databases vs Knowledge Bases

Databases

Social Insurance

Databases

Healthcare and


Social Support

Databases

Pensions

Personal data are distributed

Social Cards and their extensions

A Social card is a way of providing services to
each citizen on individual basis

Main features
:

1.
Identification of Citizen



4.
Public Transport discounts


2.
Social Benefits




5.
Loyalty programs


3.
Healthcare



6.
Payments

Samara region: The results of the First
stage of Deployment



37
towns & villages


260

Internet kiosks



Knowledge Base contains
534
laws
and regulation acts


-

278 Federal


-

164 Regional


-

92 Municipal



Works for social care, healthcare,
electricity and water supply, education
and other social domains



Social benefits for veterans, disabled
people and many other categories of
citizen



1200
00

social cards



50

Social Manager workstations



37

Knowledge Engineer workstations



6

Chief Executive Authority
workstations


Multi
-
Agent Scheduling


Java
-
based


J2EE architecture


Scalable/Robust


Strong visualizations


Desk
-
Top & Web
-
Interface


Ontologies

Enterprise

Platform

Multi
-
Agent

Technologies



Based on Semantic web technology


Ontology to capture Enterprise Knowledge
and keep it separately from source code


Decision Making Logic based on Ontology


Able to Learn (Using Pattern Discovery
module)


Swarm
-
based approach (vs mobile agents)


Supports Complex networks


Influenced by real market mechanisms


Adaptive, Real time and Event
-
driven


Agents are Pro
-
Active


Provide Emergent Intelligence


Technological Platform


Demand and Supply Matching

(orders and resources in logistics, words and semantics
in text understanding, data and clusters in Clustering)


Virtual Market

D

S

D

S

D

S

D

S

S

S

S

D

S

S

D

D

S

D

D

D

S

Demand
-
Supply

Match

Demand

Agent

Supply

Agent

Match

Contract

MAT Solutions based on Virtual Market of
Demands and Resources

MAT Solutions for Real Time Logistics





Designed for resource
scheduling in real
-
time
mode, supply chains
optimization,


business performance
enhancement

MAT Solutions for Real Time Logistics


Truck Scheduling


Ocean tankers Scheduling


Taxi Scheduling


Courier Scheduling


Car Rental Optimization


Factory Scheduling


Supply Chain Optimization

VOL: 10 PALLETS

SLA: 10 DAYS


40%

VOL: 10 PALLETS

SLA: 5 DAYS


80%

VOL: 5 PALLETS

SLA: 2 DAYS


60%

20%

20%

20%

VOL: 5 PALLETS

SLA: 8 DAYS


60%

20%

VOL: 10 PALLETS

SLA: 10 DAYS


120%

60%

60%

100%

This order has a shortest journey
route…

…but the capacity is not available
on one of the legs.

It is important to be able to assess
alternate routes, to meet services
levels and minimum cost.

Imagine the power of having a single
system that can automatically plan
and re
-
plan a network like this, as
events occur, such as new orders
being added or resource availability
changes.

Example
:
European transportation Network

Transport Logistics Network Complexity


Real
-
time scheduling with shrinking time windows


Large & complex networks (> 1000 orders per day, > 100 locations, > 50
vessels )


Less
-
than
-
Truck loads requiring effective consolidation


Need to find backhaul opportunities


Intensive use of crossdocking operations


Trailer swaps


Numerous constraints on products, locations, dock doors, vehicles: types,
availability, compatibility


Individual Service Level agreements with major clients


Own and third
-
party fleet


Fixed and flexible schedules


Dependent schedules (trailers, drivers, dock doors)


Real time economy


Activity Based Cost Model, etc


Most of large & complex networks are

still scheduled manually!

Pattern
Discovery

Resulting Plan and KPI

Adaptive
Scheduler

Events Flow

Network Designer

Ontology
Editor

Simulator

Ontology

Network (Scene)

Modeling Data

Patterns
and Ongoing
Forecast

Current Situation
and Ongoing Plan

Modeling Plan
and KPI

Domain Knowledge

Evolutional
Design

Re
-
Design of
Network

Architecture of Multi
-
Agent Platform

Multi
-
agent Scheduler: Screen Example

Truck 1

08:00

16:00

12.00

20:00

Time

Заказ

1

Order 2

Order 3


Consider a schedule


New order arrives


Preview


New order ‘wakes up’ Truck 3
agent and starts negotiations
with him


Truck 3 evaluates the options to
take New order


Truck 3 ‘wakes up’ Order 3
agent and asks it to shift to the
left


Order 3 analyzes the proposal
and rejects it


Truck 3 asks New order if it can
shift to the right


Truck 3 decides to drop Order 3
and take a New order


Order 3 starts looking for a new
allocation and finally allocates
on Truck 1 by shifting Order 1

Truck 2

Truck 3

New order

Which truck is best
for me?

I can take new order if I:


Shift Order 3 to the left


Shift New order to the right


Drop Order 3

Will you take me?

Can you shift
to the left?

I can’t shift

Can you shift to
the right?


No

Logic of Multi
-
Agent Scheduling

A

Consider logistic network of a company

1.Order1

goes from Point C to Point Z

2.Order2

goes from Point B

to Point X

3.

Заказ
3 appears
, and goes from Point A

to Point Z

4.Order3 decides to go to B and then travel with Order 2 via cross
-
dock1

5.Order4

appears and goes from Point A to Point Y

6.Order3 decides to travel the first leg with Order 4 and the second leg with Order 1 via cross
-
dock 2, to avoid going alone from A to B

Cross dock

2

Cross dock

1

B

C

Z

Y

X

Logic of Multi
-
Agent Routing

Case Study: UK Logistics Operator

Network Characteristics:


4500 orders per day


Order profile with high complexity


Many consolidations should be found


Few Full Truck Load orders


Few orders can be given away to TPC


Majority of orders require complex planning


the
price of a mistake is high


600 locations


Large number of small orders


3 cross docks


9 trailer swap locations


140
own fleet trucks, various types



20 third party carriers


Carrier availability time


Different pricing schemes

Key Problem:

Real
-
time planning
in a highly complex network with
X
-
Docks and Dynamical Routing

Problems to be Solved:


Location availability windows

Backhaul


Consolidation

Vehicle capacity

Constraint stressing

Planning in continuous mode

Dynamic routing

Cross
-
docking

Handling driver shifts

Summary of Benefits (Before / After)

BEFORE IMPLEMENTATION

AFTER IMPLEMENTATION

Two operators worked for a day

to make a schedule for 200 instructions


Planning day 1 for day 3:

no chance to

Support backhauls and consolidations

in real time


8 minutes to schedule 200 transportation

instructions

Planning day 1 for day 2 and even day 1

for day 1


No software for schedule 4000 orders

With X
-
Docks and Drivers

(manual procedure only)


Hard to consider various criteria

quickly and choose the best possible

option

4 hours to plan orders 4000 orders via

X
-
Docks and ability to add new orders

incrementally (a few seconds for a order)

Choosing the best route from the point

of view of consolidation or other criteria


Knowledge was hard to share, it was

“spread” among different experts


Capture best practice

and domain

knowledge in ontology. New

knowledge can be inserted quickly.

Key Customers


Avis (UK): Leading car rental provider


Innovative dynamic scheduling system for downtown market reducing
car assets required and improving service levels


Addison Lee (UK): largest private hire car firm in London


delivering core operational systems and dynamic scheduling


Tankers International (UK): Manage a large oil tanker fleet


development of dynamic scheduling software for shipping fleet


One Network (USA): logistics software provider


providing development services to implement new core, scheduling
and visual features/components for their platform


GIST (UK): supply chain specialist


real
-
time scheduling software tool for increased fleet utilisation and
reduced transportation costs


Enfora (USA) : major manufacturer of handheld devices


development of a wide range of software modules and market
partnership for a dynamic scheduling web service


Move forward with Multi
-
Agent Systems

That Was Then

This is Now

Batch

Optimizers

Rules Engines

Constraints

Real
-
time

Manage Trade
-
offs

Decision
-
Making Logic

Cost/value equation

Visualize

Learn,
Simulate

and Forecast

Adaptive Factory Scheduler

Main features include
:



Creation of production plans
;


Planning of production equipment, operations,
resources based on ontology;


Adaptive rescheduling in response to unexpected events
(equipment failures, operation delays, etc.);


Visualization of current production plan;


Description and updating system knowledge through
the ontology;


Semiautomatic editing of production plans. For
example, a user may change the initial plan for any
machine or equipment or add new production tasks,
change or cancel some of previous tasks and
operations, etc.


Results of scheduling are presented in
Gantt chart form showing the level and the
intensity of resources utilization in the
course of production plan fulfillment

User plans production processes by
assigning resources for their fulfillment
(machines, equipment etc.)

Adaptive Factory Scheduler

Data


knowledge
»)
about resources
are entered and stored in ontology.

Adaptive Factory Scheduler allows
operating with ontology data,
updating, modifying and deleting
them…

…and

visualize factory
ontology with adjustable
detailing level

Adaptive Factory Scheduler

Factory ontology example


In order to manufacture a driving
mirror it is necessary to make a
form, to cut glass, to paste glass to
a substrate, etc.

For this purpose we need

the
following materials: a plate, glass,
substrate, glue, and other raw
materials
.

Each operation should be carried
out by skilled worker



Actively developing in Semantic Web

for Internet pages semantic
description



Factory Ontology contains
description of basic domain objects
and

relationships between them
.



Ontology allows to represent
knowledges of certain domain
separately from program code



Ontologies usage allows to build
flexible and scalable applications
easily adopted to any business by
means of changing
«
system
knowledge
»

by demand
.



Ontologies can be successfully
applied for decision support,
learning, knowledge integration and
other areas
.

Achieved results
:



Production planning on the basis of real resources characteristics (equipment, machines,
workers), their availability at various time periods and information about changes


Combination of the planning stage with plan execution monitoring, flexible rescheduling


Monitoring of technology and production plans


More efficient strategic and tactical planning in response to maximum requirements in the
condition of uncertainty, resources distribution conflicts and high risks


Enhanced visualization capabilities (Gantt charts, semantic networks)


Higher adaptability and configuration capabilities


Execution of orders just in time through flexible planning in the real time mode

Adaptive Factory Scheduler


Solves “unsolvable” problems in complex logistic networks


Supports event
-
driven, continuous planning in real time with intelligent
reactions to unexpected events


Fast reaction: reactive and pro
-
active changes of parts of the schedule
without changing the whole schedule


Provides smart decision support and sophisticated user interaction


Reacts on events and constantly generates new options proactively


Provides individual & detailed cost calculations per order / resource


Makes trade
-
offs to balance different criteria (cost, profits and service levels)


Provides ability to override constraints


Supports collaborative team work with users


Provides integration of scheduling processes across the company


Makes decision making visual


Knowledge
-
based: Uses domain
-

and company
-
specific knowledge to
produce feasible schedules and reduce dependency on key individuals


Customizable and configurable


Platform for supporting business growth and performance increase


Reduces cost & time, improve service, lower risks and penalties


Supports
«
what
-
if
»

games for business optimization



Benefits of MAS for Real Time Logistics

KEIS: Intellectual data mining


Designed to discover patterns, hidden
dependencies and business
-
critical
knowledge in the databases, texts and other
information resources

KEIS: Intellectual data mining


Traditionally, data analysis is carried out by
human
.


However
,
human

cannot

find more than two
-
three dependencies
even in small data files
,
and at the same time mathematical
statistics operates with
averaged

parameters and cannot help in
practical

recommendation preparation
.


In contrast with the traditional methods of data analysis, KEIS
discovers
hidden

rules and dependencies
automatically
.


KEIS

is

designed

for analysis of data extracted

from different
sources and presented in different
formats
.

Problems of traditional data analysis

KEIS: Intellectual data mining

Cluster analysis basics

Clustering

is

one

of

the

basic

approaches

used

to

discover

hidden

patterns

in

the

huge

information

files


Cluster

analysis

allows

to

find

previously

unknown

dependencies

in

data
.

These

dependencies

are

hardly

discovered

using

other

approaches
.


Clustering

divides

data

into

groups

(clusters)

where

elements

inside

one

group

have

more

«
similarity
»

among

themselves

than

with

elements

in

neighbor

clusters

Clustering Technology

Data processing

Data transformation to possible
input data formats

Data loading…

Discovery of clusters

Cluster

1

Cluster

2

Cluster

3

Cluster
4

Файлы формата
txt
.
(
Блокнот
)
Файлы формата
mdb
.
(
Microsoft Access
)
Файлы формата
xls
.
(
Microsoft Excel
)
Cluster analysis

Databases



Stages of KEIS data processing

1.
Data loading

2.
Data processing

3.
Analysis by attributes

4.
Cluster analysis

5.
Cluster content analysis

6.
Automatic generation of semantic rules

Basic stages





Stage

1:
Data loading

System GUI


Data file opening

On the first stage data loading
and pre
-
processing are
executed



Pre
-
processing



Stage

2:
Data processing


Initializing clustering process

On this stage clustering of full data set
by selected attributes has to be
executed

Information about
discovered clusters
then is shown in the
table



Stage

3:
Data analysis by attributes

Detailed research of cluster
parameters using categories of
selected attribute is carried out o
n this
stage



Select

an

attribute



Stage

4:
Cluster Analysis

All

clusters

discovered

in

loaded

data

are

presented

in

pie

chart
.


Content

of

each

cluster

can

be

analyzed

in

details

in

the

system





or

exported

for

review

and

further

processing

to

Microsoft

Excel
.

This stage is a visualization stage
.
Segments of pie chart correspond with
discovered clusters, and its size allows to
evaluate number of records in certain
cluster



Stage

5:
Analysis of cluster’s content

At this stage, detailed
information about all
categorial attributes for
the selected cluster can
be presented.


Each attribute is shown
at the diagram using
colored area.


Height of this area allows
to evaluate total number
of records with certain
attribute, in selected
cluster
.

Selected

cluster

content

visualization
.




Stage

6:
Semantic rules generation

At this stage system allows to formulate
correspondences between different
attributes in a logical form
«
if

then
».

Selection

of

logical

scheme

(
condition

conclusion
)

by

the

user

Automatically generated rules are shown by
the system

generating


If user has a car, then he often travels
with:
1.
Family
; 2.
Partners
; 3.
Friends




High performance


High reliability of analysis results


Flexible cauterization parameters settings


Possibility to process
big

information files contains
hundreds of thousands of records where each record can
has hundreds of attributes


Support of
different formats

of input data

(
txt/xls/mdb
)


Possibility of clustering using many parameters


Possibility to handle both quantitative and non
-

quantitative parameters

KEIS: Intellectual data mining

Basic advantages of KEIS

KEIS: Case study
.
Social sphere


Data analysis related to recipients of social support in Kinel
town allowed to determine all groups of recipients and their
basic characteristics

Discovered clusters

Cluster diagram

KEIS: Case study
.
Car insurance


Insurance company provides car insurance service and has staff experts who on
the basis of several criteria (official requirements and personal expertise) makes
decisions of business conditions for certain client.

Guessed decisions include
providing of insurance or rejecting of service, tariffs, potential legal costs, etc.



Cluster analysis allows
«
to discover
»
hidden dependencies between client
characteristics and insurance accident risks by special client’s data processing
.

Discovered clusters

Cluster analysis allowed to find out most secure and insecure
segments of clients

KEIS: Case study
.
Mobile operator


Mobile company database analysis allows to discover main groups of
clients and their preferences

Main groups of clients
.

Different services usage s
tatistics

Cluster

№9
corresponds

to the largest segment of
clients

(6141
records



45%)

Local traffic two times more than average, roaming
is tree times less than average
,
Long distance calls
at average level,

additional services at average level
.
I
.

е.

predominantly local calls
.

Most likely
,
usual

local residents

Text Understanding with generation of
semantic network

Intellectual text processing and analysis implies understanding its
semantics
.


Text semantics can be presented in the form of a
semantic network
(scene)

-

the information structure reflecting concepts, objects,
subjects mentioned in the text, and relations between them
.


Domain ontologies are used in order to create scenes
.


Instances
:



Molecular biology article’s abstracts understanding



Insurance company contracts processing



Semantic information search



Perspective
:
Semantic
-
based terrorists SMS or e
-
mail messages (or
even phone calls) recognition

Example: Generation of semantic descriptor for
molecular biology article excerpt

Модуль построения
семантических
дескрипторов
ориентирован на анализ
реферата и создание на
основе онтологии
предметной области
семантического
дескриптора, однозначно
описывающего данный
реферат. Дескриптор для
каждого реферата
строится единожды, и
далее работа
осуществляется со
сформированной базой
дескрипторов
.


Two pUC
-
derived vectors
containing the
promoterless

xylE gene (encoding
catechol 2,3
-
dioxygenase)
of

Pseudomonas putida mt
-
2
were constructed.

The t(o)

transcriptional terminator
of phage lambda was
placed

downstream from the stop
codon of xylE. The new

vectors, pXT1 and pXT2,
contain xylE and the t(o)

terminator within a cloning
cassette which can be

excised with several
endonucleases.


Two pUC
-
derived vectors
containing the
promoterless

xylE gene (encoding
catechol 2,3
-
dioxygenase)
of

Pseudomonas putida mt
-
2
were constructed.
The t(o)

transcriptional terminator
of phage lambda was
placed

downstream from the stop
codon of xylE.

The new

vectors, pXT1 and pXT2,
contain xylE and the t(o)

terminator within a cloning
cassette which can be

excised with several
endonucleases.


Two pUC
-
derived vectors
containing the
promoterless

xylE gene (encoding
catechol 2,3
-
dioxygenase)
of

Pseudomonas putida mt
-
2
were constructed. The t(o)

transcriptional terminator
of phage lambda was
placed

downstream from the stop
codon of xylE.
The new

vectors, pXT1 and pXT2,
contain xylE and the t(o)

terminator within a cloning
cassette which can be

excised with several
endonucleases.


Analysis of the first sentence

Two pUC
-
derived vectors
containing the
promoterless

xylE gene (encoding
catechol 2,3
-
dioxygenase)
of

Pseudomonas putida mt
-
2
were constructed. The t(o)

transcriptional terminator
of phage lambda was
placed

downstream from the stop
codon of xylE. The new

vectors, pXT1 and pXT2,
contain xylE and the t(o)

terminator within a cloning
cassette which can be

excised with several
endonucleases.


Analysis of the second sentence

Analysis of the third sentence

System Architecture

Parents
-
Children

Goods



Small business

Scene

MAS for Text
Understanding

Domain Ontology

SMS
-

messages
,

е
-
mails, etc
.

Scenes
Archive

MAS for
pattern
detection

Language
Options

Patterns
Library

MAS for

scene
clustering

Signal of pattern detection

MAS for
language
queries

Clusters

Typical
Queries

Ontology extension requests

Where Vasya
waslast
week
?

Semantic network generated in course of text
analysis

Conclusion

1.
Knowledge Genesis develops innovative multi
-
agent systems
applicable to complex problems solving in various domains

2.
First experience of multi
-
agent systems development for e
-
government, adaptive planners, text understanding, clustering
demonstrates high efficiency and existence of good perspectives
of the approach on world market

3.
Currently Knowledge Genesis is working on new generation of the
high
-
performance multi
-
agent systems functioning on distributed
network of servers and allowing learning by experience

4.
We will be

happy to have new possibilities for further development
and application of our technologies in different domains to solve
complex problems

THANK YOU
!


Russia, Samara, 443001,

Sadovaya street 221

Tel/fax:
007
-
846
-
3322101

www.kg.ru

info@kg.ru