[ieee 2011 international conference on advanced technologies for communications (atc 2011) - da...
TRANSCRIPT
2011 International Conference on Advanced Technologies for Communications (ATC 2011)
Similarity based Optimization Context Awareness In
Remote Healthy Service
Thi Hien Pham, Tien Trung Do, Thi Vu Anh Ha, Luong Pham Van
School of Information and Communication Engineering
Sungkyunkwan University, South Korea
[email protected], [email protected], [email protected], [email protected]
Abstract-Context awareness service is one of the key features in ubiquitous computing system. In heterogeneous pervasive computing system, effective context modeling and reasoning are important to enable the collaboration and distributed reasoning among the agents. The effectiveness of the previous approaches for distributed reasoning significantly degrades when a large number of agents are involved. In order to solve this problem we propose a layered context model facilitating distributed reasoning. We also propose an similarity based context reasoning to optimize system performance. The performance of the proposed scheme is verified in remote health service, and it shows that the trust of reasoning outcome is greatly enhanced and the data traffic is effectively reduced.
Index Terms-context awareness; distributed reasoning; layered context modeling; multiagents system; similarity.
I. INTRODUCTION
In open heterogeneous enviromnent, context awareness is
the key of the ubiquitous system. The main challenge in
such context-aware system is to make appropriate decisions
by promptly taking into account the user's context, the data
used to characterize the user's current situation. The task of
manipulating the context data in an intelligent way is a crucial
contemporary research, which is often referred to as context
reasoning [10].
The recent research in similarity based context reasoning
tends to distributed collaboration employing multi-agent sys
tem MAS. Here two problems need to be dealt with; how to
model and reason the context.
The existing approach using Bayesian network [6, 14] or
Petri net and fuzzy logic [7] for modeling the context and
apply distributed reasoning using the agents. An existing
solution of reasoning context employs a uniform context model
to merge all the agents' perspectives in the reasoning process
[I]. Each agent holds only a partial view of the context,
therefore, the perspectives of other agents need to be merged
with its own view for proper reasoning. This becomes even
more challenging and complicated if a large number of agents
are involved. Moreover, although the geometric model use for
context-situation is used in paper [I], there is no assumption
about the relation between points, especially their geometric
distance. Since discrete states of context can be mapped to real
values, no relation can be guaranteed for neighboring points
[9].
An important requirement for context-aware service in ubiq
uitous computing enviromnent is efficient context modeling
978-1-4577-1207-4/11/$26.00 ©2011 IEEE 165
required for supporting the collaboration among the agents
considering individual perspectives. In this paper we propose
a layered context model approach in which higher level
context is inferred from lower level context models. Moreover,
based on Mahalanobis distance, the similarity based context
reasoning can take the correlation between context models
into account. Therefore, the proposed approach increases the
confidence of reasoning outcomes and maximizes the system
performance.
The rest of the paper is organized as follows. Section II
discusses the work related similarity based reasoning. Section
III introduces the proposed context modeling and similarity
based reasoning approach. Section IV deploys the proposed
context model and tests its effectiveness. Section V concludes
the paper and identifies the future work.
II. RELATED WORK
In an environment typified by complexity, openness, hetero
geneity and context uncertainty, modeling and reasoning of the
contexts in a distributed manner are essential.
Similarity based reasoning is proposed in [3, 12]. In the
proposed approaches, fuzzy logic and ontology is used for ap
proximate reasoning about the situation similarity. Fuzzy logic
is well suited for describing subjective contexts, performing
multi-sensor fusion of these subjective contexts and resolving
potential conflicts between different contexts [4]. However, in
the large scale agents system, the logic-based rules specity
become complex and difficult to gain overall picture of context
model [1].
III. PROPOSED SCHEME
In similarity based reasoning for multi-agent context aware
system, context modeling and the distribution of reasoning
process among the agents are the two key issues. Context
modeling deals with how context information is collected,
organized and represented. In this paper we propose a layered
context modeling based on multidimensional space which is
proposed in [1]. This model is proposed in the paper [13]. In
layer 1 (the lowest layer), a context model level 1 is inferred
from the original context information, while in each higher
layer, the context model is inferred from the adjacent lower
layer context without using the original context information.
A. Layered Context Modeling
• Definition 1 (Context): Context is any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves [7].
• Definition 2 (Context attribute): A context attribute ai is defined as any type of data that is used in the process of reasoning about context, ai (t) is context attribute at time t. The attribute consists of name ani and value avi respectively. Therefore, it can be represented as (ani, avD. With each value avi, we give an acceptable region Ai- Table II is an example about the relation between context attribute noise level, attribute value and acceptable region.
• Definition 3 (levell Context State): Levell context state is a set of context attributes C1 = (aI, a2, ... , an). Level 1 context modeling is inferred from original context information.
• Definition 4 (Attribute weight): Given a set of context attributes (aI , a2, ... , an ) with associate weight (Wl, W2, ... , Wn ). A weight Wi E [0, 1] represents the importance of an attribute is in describing context.
n
L Wi = 1, (1 ::.; i ::.; n), Wi E [0, 1] (1) i=1
TABLE I LEVEL T CONTEXT MODELING
Level 1 context modeling Attribute Weight Temperature 0.3
Pulse rage 0.3 The patient is taking treatment Blood pressure 0.2
Amount of Activity 0.3 Stationary duration 0.2
TABLE II CONTEXT ATTRIBUTE - PULSE RATE
Attribute name Attribute val ue Acceptable range High pulse rate 90-120 BpM
Pulse Rate Medium pulse rate 70-90 BpM Low pulse rate 50-70 BpM
• Definition 5 (contribution): Given an acceptable region A� with a contribution S of an element, reflects how well that element plays in inferring higher level context modeling. Table I describes the level I context state "professor is teaching" in studying class, in which represents relation of context level 1 modeling, context attribute, attribute weight.
• Definition 6 (Context state level (i+1)): A level i+1 f .. . CHI - (C
i C
i C
i) context state 0 source J IS. j - l,j, 2,j, ... , m,j where i � 1 to guarantee higher level context can be deduced from lower level contexts.
166
• Definition 7 (Context weight): A set of level i context states from source j: (CLj' q,j' ... , C;:,,) are used
for inferring higher level context c;+ 1. Some contexts are important than the other [15]. Given associate context weights (wctj' wctj' ... , wc';,,) are assigned for relative importance of contexts in deducing the higher level context. Table III is an example when higher level context is inferred from the lower context.
m L w4,j = 1, (1 ::.; k ::.; m), WCk,j E [0, 1] (2) k=1
B. Similarity based Context Reasoning
Although the layered context modeling is proposed in our previous work [13], there is no criterion to classify the level of the context layer. Another proposal in the paper [I] also use the geometrical space to model context-situation, there is no assumption about the relation between the adjacent points.
A fundamental requirement for autonomic computing is to be able to automatically infer how human users react in similar contextual conditions [9]. In the paper [2], the degree of similarity is defined by Euclide distance. However, the two disadvantages of the Euclidean distance are: the Euclidean distance is extremely sensitive to the scales of the variables involved and blind to correlated variables. In this paper we propose Mahalanobis Distance to obtain the similarity of context models by taking the correlation between context models into account. Therefore, the confidence of reasoning process is increased.
• Definition 8 (The lvI ahalanobis Distance between context models): Given two context models CJ and
CrThe dissimilarity- Mahalanobis Distance measure of the same distribution with covariance matrix Sis:
where m is the total number of unique context attributes and qi is the sum of weight of overlap region between attribute .
• Definition 9 (The Degree of Similarity between two context models): The Degree of Similarity (DoS) between two context models CJ and Clk is:
DoSj1 = 1 - Dif fJf (4)
• Definition 10 (total Degree of Similarity): Given a set of known context models level k: Ch i = 1,2, ... , n). For a unknown context model leveli: Cj (i # k, I # j), the total Degree of Similarity between Cj and set of Clk
models (I = 1,2, ... , n) is:
m DoSj1 = L(1- Dif fJf)
1=1
IV. PERFORMANCE EVALUATION
(5)
In this section, we implement two simulations. In the first simulation is similarity-based context reasoning. Here Mahalanobis distance and Euclidian distance are compared.
TABLE III HIGHER LEVEL CONTEXT INFERED FROM LOWER LEVEL CONTEXT
Level 3 context modeling Level 2 context modeling wcL Level 1 context modeling wci
The patient is beling followed 0.2 The patient is sleeping 0.4
Patient's checking room The patient is doing exercise 0.6
The patient is being periodically healthy checked 0.2 The patient is taking a treatment 0.7
The doctor is checking 0.3
The patient is being in emergency situation 0.6 The patient is in sudden changing condition 0.5
The patient is in sudden changing activity 0.5
The system tries to adapt the new context by searching the distance is useful for predicting whether he or she has any
most similar model in the known context models. The second disease or not.
simulation is the effectiveness of proposed scheme in reducing
the data traffic.
Fig. 1. The architecture of context aware system.
A. Agent-based Context Aware System
The agent technology has been widely accepted as an
efficient paradigm for developing intelligent services when
distributed computing environment [8, 15]. In this paper we
implement a healthy remote service employ multi-agent con
text aware system (figure 1). The system consists of context
aware agent, aggregator, and service agent. In the low level,
the agent aggregator gather the context attributes by RF-ID
reader and then disperse them to infer the higher level context
model.
B. Similarity-based context reasoning
1) Degree of Similarity between People: In this part, we
assume there is a new person (Table I) who comes to the
hospital to check the healthy. Then, we can infer the confi
dence of Mahalanobis distance and Euclidean distance when
they are used to predict the conditions of the patient. Note
that the criterions to define whether a person has any disease
or not was predetermined before. The first simulation tests a
normal person while the second simulation checks the diseased
person. In both of cases, Figure 2 and Figure 3 show that the
confidence obtained by Mahalanobis distance is much higher
than that of Euclidean distance. Therefore, the Mahalanobis
167
� SJ! 'E OJ '5 m
� m
'"
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
1.0
0.9
0.8
0.7
..,. � 0.6
'E en 0.5 '5 � 0.4
al' o 0.3
0.2
0.1
0.0
0
---T- Euclidean Distance of Normal Person
� Mahalanobis Distance of Normal Person _ Euclidean Distance of Diseased Person -+- Mahalanobis Distance of Diseased Person
20 40 60 80 100
Time(ms) 120 140
Fig. 2. The predicted result of a normal person.
....
o
..... �
--T- Euclide:; Dislance of Normal Person
I � Mahalanobis Distance of Normal Person
_ Euclidean Distance of Diseased Person -+- Mahalanobis Distance of Diseased Person
.A'- '""-...,...
20 40
--
60 80 100
Time (ms)
�
120 140
Fig. 3. The predicted result of a diseased person.
2) Resource Consumption: In ubiquitous environment, the
context aware system updates information when the condition
of the environment changes. Therefore, the agent needs to
communicate with the other by exchanging the messages
between them. In this paper, we reference the hierarchical
P2P networking scheme which is proposed in the paper [16].
TABLE IV THE NORMAL AND DISEASED PERSON
Temperature (Celsius) Pulse rate (BmP) Blood pressure (mmHg) Amount of activity 36.5 - 37.5 80-120 38- 40 110-130
Here the sender agent compares the message with the previous
one before sending it. If the similarity between the messages
is high (Figure 4), the agent does not send the message
but saves it in the local store. This case happens because
of: the new context model is quite similar to the known
model (Mahalanobis distance based). As a result, the proposed
scheme decreases the amount of traffic when the context data
does not change frequently (Figure 5).
•
Compere
Fig. 4. Message composition in the sender agent.
4000 � Proposed Scheme _Scheme (1)
3500
3000
_ 2500
� � 2000
� � 1500 o
1000
80
Time(ms)
100 120 140
Fig. 5. Throughput of the proposed scheme compare to the scheme in (1).
V. CONCLUSION
In this paper, we propose a novel layered context model
in which higher level context model can be inferred from
lower level context models without any requirement of original
context information. The Mahanobis distance is used to define
the similarity between the unknown context model and the
known context model. The system tries to optimally adapt the
new model by finding the most similar known context model.
168
110-140 250-450 130-160 100-200
The proposed scheme is tested by computer simulations
to prove that our approach enhances the trust of reasoning
outcomes than the earlier approach. Moreover, a large amount
of data traffic can be saved in adapting two similar context
models. As a result, the proposed approach optimize the
system performance.
In the future work, we will research on how to apply
distributed reasoning autonomously to enhance personalize
intelligent service. We will extend the context model and
context adaption in stability and reliability for adapting context
aware pervasive environment.
REFERENCES
[1] Amir Padovitz, Seng W. Loke, and Arkady Zaslavsky, "Multiple-Agent Perspectives in Reasoning About Situations for Context-Aware Pervasive Computing Systems," Systems, Man and Cybernetics, Part A, IEEE Transactions on 38(4): 729-742, 2008.
[2] Amir Padovitz, Seng Wai Loke and Arkady Zaslavsky, "Toward a Theory of Context Spaces," Workshop on Context model and Reasoning (CoMoRea), ,IEEE, Florida, pp. 38-42, March. 2004.
[3] Christos B. Anagnotopoulos, Yiorgos Ntarladismas, and Stathes Hadjietlhy miades, "Reasoning about Situation Similarity, " Intelligent Systems, pp. 109 - 114, IEEE, Sept. 2006.
[4] Claudio Bettini, Oliver Brdiczka, Karen Henricksen, "A survey of context model and reasoning techniques," Pervasive and Mobile Computing, Volume 6, pp. 161-180, April. 2010.
[5] Dey, A. K, 'Understanding and Using context', Personal and Ubiquitous Computing archive, Volume 5, Issue 1, February. 2001.
[6] Gregor Pavlin, Patrick de Oude, Marinus Maris, Jan Nunnink and Thomas Hood, "A multi-agent systems approach to distributed bayesian information fusion," Information fusion , Volume 11, Issue 3, pp. 267-282,
July. 2010. [7] .Tian Ye, Jintao Li, "Ubiquitous Computing-Oriented Distributed Fuzzy
Reasoning Petri Net Modeling and Simulation," International Conference on Parallel and Distributed Computing, Applications and Technologies 2009, IEEE, Higashi Hiroshima, pp. 224 - 230, Dec. 2009
[8] Joon Hwang Kim, Jeong Hun Chu, Seungwok Han, Chang Won Park and Hee Yong Youn, "Hierarchical P2P Networking and Two-level Compression Scheme for Multiagent System Supporting Context-aware Applications," Proceeding of ISDA 2008, IEEE, Kaohsiung, pp. 519-523, Nov. 2008.
[9] Nearchos Paspallis, Konstantinos Kakousis, George A. Papadopoulos, "A multi-dimensional Model Enabling Autonomic Reasoning for Contextaware Pervasive Applications, " Mobile Ubiquitous 2008.
[10] Petteri Nurmi, Patrik Floren , "Reasoning in Context-Aware Systems". [II] P.C.Mahalanobis, "On the generalized distance in statistics, "Jan. 1936. [12] Swapan Raha, Nikhil Ranjan Pal, and Kumar Sankar Ray, "Similarity -
Based Approximate Reasoning: Methodology and Application, " Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Volume 32, pp 541-547, Jul. 2002.
[13] Thi Hien Pham, JunYeol Choi, Hyung Do and Hee Yong Youn, "Distributed Multi-agent Reasoning with Layered Context Modeling and Priority", CPS conference, Hangzhou, IEEE, pp.677-684, 2010.
[14] Yang Xiang, "Distributed Multi-agent Probabilistic Reasoning with Bayesian Networks," Lecture Notes in Computer Science, Springer Berlin, Jan. 2006.
[15] Z. Wei, N. Wang, M. Kang and W. Zhou, "An agent-based contextaware middleware for pervasive computing," Information Science and Engieering 2008, IEEE, Shanghai, pp. 116-119, Dec. 2008.