Elsevier

Neurocomputing

Volume 135, 5 July 2014, Pages 86-97
Neurocomputing

A multi-agent system for enabling collaborative situation awareness via position-based stigmergy and neuro-fuzzy learning

https://doi.org/10.1016/j.neucom.2013.03.066Get rights and content

Abstract

Situation awareness is a computing paradigm which allows applications to sense parameters in the environment, comprehend their meaning and project their status in the next future. In collaborative situation awareness, a challenging area in the field of Ambient Intelligence applications, situation patterns emerge from users' collective behavior. In this paper we introduce a multi-agent system that exploits positioning information coming from mobile devices to detect the occurrence of user's situations related to social events. In the functional view of the system, the first level of information processing is managed by marking agents which leave marks in the environment in correspondence to the users' positions. The accumulation of marks enables a stigmergic cooperation mechanism, generating short-term memory structures in the local environment. Information provided by such structures is granulated by event agents which associate a certainty degree with each event. Finally, an inference level, managed by situation agents, deduces user situations from the underlying events by exploiting fuzzy rules whose parameters are generated automatically by a neuro-fuzzy approach. Fuzziness allows the system to cope with the uncertainty of the events. In the architectural view of the system, we adopt semantic web standards to guarantee structural interoperability in an open application environment. The system has been tested on different real-world scenarios to show the effectiveness of the proposed approach.

Introduction

Situation awareness is a computing paradigm that enables applications to sense and explore situations in which the users are, with the aim of predicting their demands at a certain time [1]. The paradigm relies on the context, that is, all the relevant data and information (e.g., the user's position in space and time, the surrounding things and events) which can help comprehending what is happening in the environment [2], [3], [4]. This form of autonomous perception implies reasoning, decision, adaptation, and other characters of cognitive systems [5], as well as dealing with an intrinsic uncertainty in data [6], [7].

To this aim, Korpipää et al. [8] have proposed a framework for managing uncertainty in raw data and inferring higher-level context abstractions with a related probability. Fuzzy sets are employed to convert unstructured raw data into a representation defined in a context ontology through predefined fuzzy labels. Situations are recognized by means of a basic Bayes classifier, which learns conditional probabilities from training data for each situation. In [9] fuzzy quantization is used to convert raw sensor data into context information. Such information is exploited by fuzzy controllers for adapting applications to the specific context. However, no semantic description of context is considered. Ranganathan et al. [10] have modeled uncertainty in situation awareness by associating a confidence value with all pieces of contextual information. The authors adopt three methods to infer the user's situation: (i) probabilistic logic, (ii) fuzzy logic, and (iii) Bayesian networks.

In [11] uncertainty is managed by first extending the context ontology so as to allow additional probabilistic markups and then by adopting Bayesian networks to infer the current situation of the user. In [12] contextual information is codified in the antecedent part of linguistic rules whose consequent parts express the degree of confidence in the occurrence of a situation. Weights can be specified to represent the relative importance of each contextual condition for inferring a situation. In [13] a neuro-fuzzy classification system is trained to map sets of contextual information to particular situations by fuzzy rules.

In [7], [14] we have proposed a design method for managing situation awareness. This method is based on the concurrent use of a semantic and a fuzzy engine. The semantic engine can infer one or more situations exploiting symbolic information. When multiple situations are inferred, a fuzzy engine computes a certainty degree for each situation, taking the intrinsic vagueness of some conditions of the semantic rules into account.

The structure of rules has been designed according to an upper situation ontology which is domain independent. The user calendar acts as a reference for the parameterization of such fuzzy rules for each user. The use of a calendar is however an explicit input required to the user. On the contrary, context information should be collected in terms of collaborative implicit input, coming from changes in the environment.

To avoid using explicit inputs as context sources, in [15], [16] we have proposed an approach based on the emergent paradigm [5] for automatically detecting social events (e.g., meetings, conferences, festivals, entertainment, and so on) by exploiting a position-based stigmergy paradigm. Stigmergy can be defined as an indirect communication mechanism that allows simple entities to structure their activities through the local environment [17]. The approach has been referred to as collaborative situation awareness (CSA). In particular, each user is associated with one marking agent which leaves periodically marks in the environment in correspondence to his position. In a stigmergic computing scheme, the environment acts as a common shared service for all entities enabling a robust and self-coordinating mechanism. The accumulation of these marks is monitored by an event agent which detects events based on a fuzzy information granulation process. Finally, situation agents infer user situations from the underlying events. The inference process is performed by fuzzy rules generated by an expert taking some mathematical constraints into consideration.

In this paper, we extend our approach by focusing on a multi-agent architecture. Further, in order to make the approach completely independent of the user's inputs, we generate the fuzzy rules by exploiting a neuro-fuzzy system. We adopt Gaussian membership functions and train the neuro-fuzzy system by tracing a number of users involved in a social event. We need only to know the number of users who participate in the event. The proposed scheme is tested on four representative real scenarios, considering four different types of situation. For each scenario, the scheme has proved to be able to recognize the four types of situation just approximately at the instants when these situations occur.

The paper is organized as follows. In Section 2, we introduce the functional view of the system. Section 3 shows the architecture of the system, by focusing on the knowledge representation. In Section 4, we discuss some experimental results. Section 5 draws some final conclusion.

Section snippets

The functional view of the system

Situation awareness is achieved in our multi-agent system by exploiting three processing levels: the marking, the fuzzy granulation and the inference processing levels. In this section, we will describe how the three levels work and interact with each other. The first two levels will be discussed shortly. The interested reader can refer to our previous paper [16] for details. The third level will be analyzed in depth. Indeed, unlike in [16], where fuzzy partitions were generated heuristically,

Architecture and knowledge representation in the CSA system

A robust and general approach to CSA should guarantee that system architecture and behavioral knowledge can be easily integrated in an open environment. Further, a variety of contextual, possibly uncertain, collective inputs should be supported. Finally, situational knowledge should be provided to multiple applications. To this aim, the architecture of the system has been designed in compliance with an agent-oriented approach [20], [21], [22], which operates at the knowledge level, shows

Simulation results

To assess the effectiveness of the proposed multi-agent system in detecting collaboration situations, we tested our scheme on four real-world scenarios involving a different number U of participants (P1,,PU). The four scenarios, denoted as A, B, C and D, refer to a meeting among 10, 8, 6 and 4 participants, respectively. Scenarios are characterized as follows:

  • Scenario A (U=10). P1 meets P2 at a bar before arriving at the meeting place. P8 reaches P1 and P2 at the bar and then together they go

Conclusions and future work

In this paper, we presented a multi-agent system for the detection of situations related to social events via position-based stigmergy and neuro-fuzzy learning. The proposed system is structured into three different processing levels managed by different agents in order to recognize situations through inference of fuzzy rules. Antecedent and consequent parameters of fuzzy rules are automatically defined by means of neuro-fuzzy learning. The system was tested on real-world meeting scenarios

Giovanna Castellano received the Laurea degree in Computer Science from the University of Bari, Italy, in 1993. From 1993 to 1995 she was a fellow researcher at the Institute for Signal and Image Processing (CNRBari). In 2001 she received the Ph.D. in Computer Science from the Department of Computer Science at the University of Bari, Italy, where she became Assistant Professor in 2002. Her research interests fall in the area of Computational Intelligence, with special focus on artificial neural

References (29)

  • A. Ciaramella et al.

    A situation-aware resource recommender based on fuzzy and semantic web rules

    Int. J. Uncertain. Fuzziness Knowl. Based Syst.

    (2010)
  • P. Korpipää et al.

    Managing context information in mobile devices

    IEEE Pervasive Comput.

    (2003)
  • A. Ranganathan et al.

    Reasoning about uncertain contexts in pervasive computing environments

    IEEE Pervasive Comput.

    (2004)
  • T. Gu, H.K. Pung, D.Q. Zhang, H.K. Pung, D.Q. Zhang, A Bayesian approach for dealing with uncertain contexts, in: 2004...
  • Cited by (12)

    View all citing articles on Scopus

    Giovanna Castellano received the Laurea degree in Computer Science from the University of Bari, Italy, in 1993. From 1993 to 1995 she was a fellow researcher at the Institute for Signal and Image Processing (CNRBari). In 2001 she received the Ph.D. in Computer Science from the Department of Computer Science at the University of Bari, Italy, where she became Assistant Professor in 2002. Her research interests fall in the area of Computational Intelligence, with special focus on artificial neural networks, fuzzy systems, neuro-fuzzy modeling, granular computing, web personalization and fuzzy image processing. Within these research areas, she has been author of more than 160 papers published on scientific journals, books and conference proceedings. Currently, she serves as associate editor of Information Sciences and she is on the Editorial Board of two international journals.

    Mario G.C.A. Cimino received the Ph.D. degree in Information Engineering from the University of Pisa, Italy, in 2007. He was a six-months visiting Ph.D. student at the Electrical and Computer Engineering Research Facility of the University of Alberta, Canada. He co-organized three editions of the Workshop on Computational Intelligence for Personalization in Web Content and Service Delivery. From 2008 to 2012, he was a Postdoctoral Research Fellow at the Department of Information Engineering of the University of Pisa, where he is currently a Researcher in Information Systems since June 2012. He is (co-) author of about 30 publications.

    Anna Maria Fanelli is a Full Professor of Computer Science at the Department of Computer Science of the University of Bari, Italy. From 2005 to 2009 she was a Director of the Ph.D. School in Computer Science at the University of Bari. Currently, she is a Director of the Computer Science Department at the University of Bari and a chair of the CILab (Computational Intelligence Laboratory) at the same department. Her recent research interest focuses on the analysis, synthesis and applications of Computational Intelligence techniques with emphasis on the interpretability of fuzzy rule-based classifiers and Web Intelligence. She is co-author of more than 200 papers in international journals, books and conference proceedings. She is co-editor of international books on the above topics.

    Beatrice Lazzerini is a Full Professor at the Department of Information Engineering of the University of Pisa, Italy, where she teaches “Computational Intelligence”, “Intelligent Systems” and “Decision Support Intelligent Systems”. Her research focus lies in the area of Computational Intelligence, with particular emphasis on fuzzy systems, neural networks and evolutionary computation. She has co-authored seven books and has published over 180 papers in international journals and conferences. She was involved in several national and international research projects. From 2003 to 2010 she was a President of the Specialized Laurea Degree in Computer Engineering for Enterprise Management of the University of Pisa.

    Francesco Marcelloni received the Laurea degree in Electronics Engineering and the Ph.D. degree in Computer Engineering from the University of Pisa in 1991 and 1996, respectively. He is currently an Associate Professor at the University of Pisa. His main research interests include situation-aware service recommenders, multi-objective evolutionary fuzzy systems, energy-efficient data compression and aggregation in wireless sensor nodes, and monitoring systems for energy efficiency in buildings. He has co-edited three volumes, three journal special issues, and is (co-)author of a book and of more than 170 papers in international journals, books and conference proceedings. Currently, he serves as an associate editor of Information Sciences and is on the Editorial Board of other international journals.

    Maria Alessandra Torsello received in 2008 the Ph.D. degree in Computer Science from the Department of Computer Science at the University of Bari, Italy, defending a thesis titled “Web Intelligence: a Neuro-Fuzzy Web Personalization System”. Her activity research mainly concerns the field of Computational Intelligence and the design of self-adaptive learning systems with particular interest in artificial neural networks, fuzzy logic, neuro-fuzzy modeling, knowledge-based neurocomputing, Computational Web Intelligence. She participated as a speaker to several International conferences. She co-organized three editions of the Workshop on Computational Intelligence for Personalization in Web Content and Service Delivery. She is co-author of about 40 publications.

    View full text