Report April_2004 - CReWMaN

brewerobstructionΤεχνίτη Νοημοσύνη και Ρομποτική

7 Νοε 2013 (πριν από 3 χρόνια και 9 μήνες)

62 εμφανίσεις

PSI Project:

Pervasive Computing Group

(Mohan Kumar, Behrooz Shirazi, and Hitha Alex )


The goal of PSI problem is to design and implement a framework for disaster
management that will monitor, prevent and recover from natural, unexpected and
inflicted di
sasters. In other words, the framework should acquire information from all
possible sources in real time, analyze and interpret this information,
detect

a crisis
situation
automatically and
as quickly as possible and take necessary action to prevent or
re
cover from the crisis situation.
An

overview of PSI framework is given in figure 1.

Crisis Identification, which is the a
utomated

and intelligent detection of security
e
vents in PSI framework may involve the following steps:


1.

Utilizing a thorough
Knowle
dge Base

2.

Information Collection from various sources

3.

Process and Normalize Information

4.

Information Fusion

5.

Identify security events

6.

Store data for later use and processing.

7.

Learn from the identified situation and update the knowledge base


PICO


The basic c
omponents in PICO are camileuns, delegents and communities


Camileuns
:


These are a
bstract representation of different hardware units
. A camileun is
represented ad


Update
s

Data

Update
s

Data

Crises

INFORMATION
COLLECTION

(Low level data translation
and

low lev
el

data
fusion,
data storage)

CRISIS
IDENTIFICATION

(Interpretation,
data

fusion
models, crisis prediction
models,

Learning
)

CRISIS
RESOLUTION

(
Situation aware
course

of action
,
Learning)

Knowledge Base

Sensor networks, news, emails,
web sites, videos
and images,
databases, applications, human
inputs

Knowledge Base,
human inputs as
the crisis unfolds

Figure 1: A
sub
-
problems overview of PSI framework

C = <H, F>, where H is the system characteristics and F is the set of
Functionalities


Del
egents
:


These are s
oftware elements that perform a single or a set of specific activity
.
These are also the a
tomic entities

that are

responsible for the creation and
operation of the communities
. A delegent can be represented as


D = <M, R, S> where M is
a set of modules that make up the delegent, R is the set
of operational rules among these modules and S is a set of services that can be
provided by the delegents


Communities

A community is a

logical organization of one or more delegents to accomplish a
single or set of goals that cannot be achieved by the delegents independently.

A community, P is given by


P = <U, G, E>, where U is the set of delegents, G is the set of goals and E is the
set of operational characteristics




Types of delegents identifie
d for PSI


There are 4 kinds of delegents that are identified for PSI:


Function

Deals with

Processes

Perception

Explicit data, raw data, its
states and values

Sensing, Detection, (Data
storage)

Comprehension

Implicit Meanings,
Situation types

Interpreta
tion, Synthesis

(Data Mining, pattern
recognition)

Projection

Future Scenarios, possible
outcomes

Prediction, Simulation

Resolution

Intentions, course of
action

Decision


浡歩湧Ⱐ
m污湮楮l


Projection Delegents for automated detection of security events
:


1.

Knowledge Base

2.

Information Collection from various sources

3.

Process and Normalize Information

4.

Correlation, Data Mining, Pattern Recognition

5.

Decision Making

a.

Identify a positive or negative security situation for example

6.

Store data for later use and proce
ssing and update knowledge base.


Steps 3, 4 and 5 are collectively referred to as
information fusion
.


Challenges

in

Information Fusion


1.

Heterogeneity

of sensors, information granularity and modes

2.

Redundancy

a.

Many sources of information about the same feat
ures

3.

Complimentary Information

a.

Sources of information about many features of the same object

4.

Timely Information

a.

Information from same sources at different instance of time

b.

Dynamic changes in the information sources/inputs

5.

Time and Cost of

information

acqu
isition

6.

Different levels of uncertainty

7.

Dynamic changes in the information sources/inputs

8.

How to decide which information to choose at any particular time as the best
information may change with time and situations

a.

sufficiency and efficiency


Information f
usion Challenges:
Dealing with
Uncertainty


Information about real
-
life security problems is rarely known with complete
confidence.

A security event may be the result of a combination of n number of reasons.
These reasons and number may not be the same for

the same security event at two
different occasions. Conclusion that a security event has happened from the collected
information may be made with only a degree of certainty in most cases.

Fundamental
bottleneck in the construction of an automated framewor
k is there for UNCERTAINTY
.

There are many techniques in information fusion that deals with uncertainity. Examples
include B
ayesian approach
,
Dempster
-

Shafer Theory

,
Fuzzy Set Theory

and
Neural

Networks
.


Bayesian Networks

Bayesian networks are complex
diagrams that organize the body of knowledge in
any area by mapping out a cause
-
and
-
effect relationships among key variables and
encoding them with numbers that represent the extent to which one variable to likely to
affect another.

These are
Directed
Acy
clic
graphical
(
DAGs)
representation of joint
probability distributions for a large number of random variables.
These are a
lso known as
belief networks.

Nodes represent

the variables and edged represents the conditional
dependencies between the variables
.





Advantages of using Bayesian Networks
:


1.

Unified hierarchical probabilistic model for information representation, integration
and inference
.

2.

The uncertainties associated with the evidential information can also be incorporated
.

probabilistically in each

node as an attribute to characterize the quality of the
information
.

3.

Evolve and grow to accommodate new events

4.

Can predict the influence of possible future actions on current tasks and vice versa

5.

Account for Temporal changes

( Dynamic Bayesian Networks)

6.

A
llows
Selection and
Decision Making

a.

C
hoose the action considering

i.

The probability

of the
occurrence

of the security situation given the
evidence at the given time,
P(H|E)

ii.

Cost/benefit of the action ( utility function in decision theory)

1.

Expected utility de
pends on the current available sensor data,
internal knowledge and current goal.


Information fusion challenge: Meeting sufficiency and efficiency


Since the best information availbale changes with time, an active selection of information
sources may be de
sired in time critical applications that has complimentary information
sources available. A framework for an active information fusion is depicted in the Figure
below:






Information Fusion

Get Best

Information

Final Decision

Active Selection


Confiden
ce

Measure
?

E1

E2

E3

En

sensors

yes

no




Projection delegent with active information fusion


D = <M, R, S>


Here M is the set of modules that
make up a projection delegent given by
[M = {m1,
m2…, mn}]

.


S is the service and in the case of a projection delegent, crisis identification is the service.


R is the rule set that defines the interaction
s between modules for achieving the services

and can be given by:




Bayesian networks, will part of this rule set in the case of a projection delegent.


j
k
i
j
k
i
m
EVENT
m
r
m
m
R
r
"
"
|
,
,