A Framework to Enable Two-Layer Inference for Ambient Intelligence Xiang Su, Davide Fucci and Jukka Riekki
Abstract In this paper, we present a two-layer inference framework to enable Semantic Web technology-based intelligent functionalities in ambient environments. The basic idea is that the low level inference is performed on the mobile devices capable of utilizing ontology, and only the high level inference is performed at the server side. This paper presents the design of this framework and illustrates its usability by a use case. The framework fully utilizes the computing capabilities of devices in the system and this way minimizes the communication among devices.
1 Introduction Ambient environments will contain plentiful of devices with different capabilities in our surroundings and our lives. Some devices are resource-constrained, for example sensors in sensor networks have limited processing power, bandwidth, and memory. On the other hand, powerful mobile terminals can even contain knowledge-based systems. Therefore, it’s a challenge to utilize the capabilities of devices fully to enable intelligent functionalities in the environment. Semantic Web technologies offer good opportunities to tackle this challenge. Our aim is to utilize these technologies to build interoperable solutions for connecting the devices in the ambient environments and using the full capabilities of each device in offering a rich set of services for the users. Xiang Su Intelligent Systems Group and Infotech Oulu, University of Oulu, Finland, e-mail:
[email protected] Davide Fucci Dipartimento di Informatica, Universit`a degli Studi di Bari, Italy e-mail:
[email protected] Jukka Riekki Intelligent Systems Group and Infotech Oulu, University of Oulu, Finland, e-mail:
[email protected]
1
2
Xiang Su, Davide Fucci and Jukka Riekki
In this paper, we are proposing a framework to enable a two-layer inference according to the capabilities of different devices. In this framework, resourceconstrained devices deliver measurement data to servers and mobile devices. Ontologyenabled mobile devices perform simple inference tasks (low-level inference) based on the data from sensors. We define an ontology-enabled device as a device capable of utilizing ontology to model the application domain and to reason about the properties of the application domain. The server performs high-level inference, including all complex tasks like advanced inference, context extraction and reuse, coordinating the devices, etc. This framework enables harnessing the computing capabilities of the system. The resource-rich servers support the Semantic Web based intelligent functionalities. On the other hand, some inference can be deployed to less capable devices like mobile terminals. For example, low level inference engine can be deployed on mobile devices to perform RDF [9] inference and simple OWL-based inference [4]. When inference is performed on ontology-enabled mobile devices there is no need to transfer sensory data, user profiles etc. to the server side. That is, the framework facilitates minimizing the amount of communication. This is an important prerequisite for the creation of a general framework for all the possible usages, because in many practical applications, capabilities of sensors and communication links in the system are limited. The idea of a two-layer approach for inference has been reported by [1] and [6]. They addressed specific problems like building the semantic representation of the user profile and emotional status, but did not consider any general framework. We expect to develop a more general and extensible framework for supporting a wider set of devices and application scenarios. Besides, some general ambient reasoning solutions have been discussed before. For example, [5] proposed a context-aware architecture supporting context reasoning, [11] presented an intelligent system based on low level, distributed, fuzzy inference learning and [2] proposed a solution integrating reasoner into a middleware. Some of them only focused on high-level inference, while others focused on low-level. None of them considered the different capabilities of devices, not to mention communication. In our framework, we will consider harnessing the full capabilities ambient systems. We will continue this article with introducing standards and technologies in section 2. Details of the design considerations of the framework will be presented in section 3, and a use case will be presented in section 4. Finally, we will draw conclusions in section 5.
2 Standards and Technologies To enable a flexible, extensible, and effective intelligent framework, we are using standardized Semantic Web technologies, like RDF, OWL, and SWRL [7] rulebased inference engine. Besides, we are employing Entity Notation (EN), which
A Framework to Enable Two-Layer Inference for Ambient Intelligence
3
is a lightweight data/knowledge representation for communication in the system innovated by [10]. It can support Semantic Web-based inference straightforwardly. RDF (Resource Description Framework) is currently the standard for establishing semantic interoperable systems. It facilitates interoperation because it provides a data model that can be extended to address ontology representation, and it enables the automatic processing for resources. In our framework, RDF is utilized as a low level representation for expressing knowledge. For example, sensory data will be transformed into RDF statements as the first step for the semantic processing. We encode RDF triplets using RDF/XML, which has been proposed by [3] as a standard to encode RDF models. To represent the ontology layer knowledge, the W3C suggests OWL as a standard [8]. OWL ontology consists of a set of axioms which place constraints on sets of individuals and the types of relationships permitted between them. These axioms provide semantics by allowing systems to infer additional knowledge based on the data provided explicitly. In this framework, we are using OWL as the main representation for context modelling as the basis for the inference. Both high-level and low-level inferences need ontology in the reasoner. Rule-based inference is utilized to derive new knowledge from the assertions in ontology at the server side. We adopt SWRL rules; a rule presents an implication between an antecedent and a consequent so that the intended meaning is: whenever the conditions specified in the antecedent hold, then the conditions specified in the consequent must also hold. The atoms composing both parts can be expressed in the form C(x), P(x,y), sameAs(x,y) or differentFrom(x,y), where C is an OWL description, P is an OWL property, and x,y are either variables, OWL individuals or OWL data values. Entity Notation is proposed as a lightweight representation for transforming information between resource-constrained devices and knowledge-based intelligent systems. This representation can be handled by resource-constrained devices, transferred over modest communication links, and transformed into knowledge (RDF and ontology) in a straightforward fashion. It’s a practical data representation with expressive power. We utilize EN as the main representation for data and knowledge exchange in this framework.
3 Designing the Framework The framework we are proposing depicts a two-layer structure that enables different inference functionalities for different devices. Figure 1 shows a concept diagram of our framework, which includes an ontology (consisting of T-Box and A-Box) and a rule set. The T-Box includes assertions on concepts, while the A-Box includes assertions on individuals, which are RDF statements. In the T-Box, all concepts and relations from the environment are described as one part of context, even though we do not give an overall concept hierarchy in this picture.
4
Xiang Su, Davide Fucci and Jukka Riekki
EN facilitates the transformation from different data sources (sensor data, user profiles, etc.) to RDF statements, and RDF statements can be mapped to the concepts in the T-Box unambiguously. The low-level inference makes use of A-Box level and small parts from T-Box level. The usage of T-Box level knowledge enables low level inference to have stronger capability than simple RDF inference, because of the description logic inference from the T-Box. But the low level cannot support any designer-specific rule-based inference. High level inference engine can support the full capabilities of ontological inference and rule-based inference. Complex intelligent functionalities, like multiple user interactions and device interoperability, can be realized at high level inference.
Fig. 1 Concept Diagram of Two-Layer Inference Framework
Figure 2 illustrates how the conceptual diagram can be deployed at devices in ambient environments. A server, ontology-enabled mobile devices, and resourceconstrained sensor nodes form of a two-layer inference system. Knowledge-based systems manage domain ontology at the server side, which host all advanced knowledge-based applications. All Semantic Web functionalities, like reasoning, publishing semantic data and semantic matching, can be implemented at the server side. Mobile devices share parts of domain ontology from the server and perform low-level inference based on shared ontology and RDF statements. Some other devices, like sensors, have really limited resources. They can send data to the nearest intelligent mobile devices. The data can then be forwarded to the server, when necessary. Servers and mobile devices can send commands to some sensors as well. The data flow of a specific system can be decided at the design time. Data is delivered as
A Framework to Enable Two-Layer Inference for Ambient Intelligence
5
EN packets. Generally, sensor data for realizing self-adaptive functions of devices can be sent to ontology-enabled devices, while sensor data for realizing complex functions, like multi-user interactions and rule-based inference, should be sent to a server. Entity Notation plays an important role in the communication of this framework. A knowledge-based system on a server can compose EN packets to transfer parts of the ontology to ontology-enabled mobile devices. Then these devices decompose EN packets and get the knowledge they need in their intelligent applications. Ontology-enabled mobile devices can also use EN composer to compose EN packets and transfer them to knowledge-based systems on servers. Resource-constrained devices cannot handle any ontology or reasoner. But still, they can compose EN packets based on their measurements and transfer them to ontology-enabled mobile devices or servers. Knowledge-based systems can reason on the transferred data and offer semantic support for these resource-constrained devices. More details about EN can be found from the paper [10].
Fig. 2 Framework of two-layer inference in ambient environments
6
Xiang Su, Davide Fucci and Jukka Riekki
4 A Use Case In this section we are presenting a use case to illustrate the usability of our framework. This example demonstrates how data is passed among main components, and how simple inference and complex inference is achieved. A social application scenario is as follows. A couple is driving home from different working places after working, and one of them needs to pick up their children from the kindergarten. The couple has decided to pick up their children as early as possible. When a parent has left the workplace she/he is reminded to pick up the children if she/he is the nearest parent already left the work. At the same time, another reminder will be shown to suggest the spouse to drive home directly.
Devices to enable the two-layer inference for this scenario include: simple GPS receivers for children, a server and ontology-enabled mobile devices with RFID readers, GPS receivers and map applications for parents. When a parent gets off work, he touches with his phone an RFID tag nearby the entrance of the building. An EN packet like this records the time he gets off: [urn:uuid:739ae5 "EntranceB" "JohnSmith" "2010-03-25T16:21:40"]
In this packet, urn:uuid: 739ae5 determines the template for this packet and facilitates the transformation from an EN packet to a RDF statement. Here is the corresponding RDF statement of this EN packet. We utilize xmlns:e=“http://ee.oulu.fi/o#” in this statement. ExitPlace JohnSmith 2010-03-25T16:21:40
Low level inference can be performed based on this RDF statement and an ontology. In the following ontology snippet, a person who leaves the building after 3 p.m. is assumed to go home. The embedded reasoner on ontology-enabled device deduces that John Smith is going home, and starts to send GPS data to the server. 15:00:00
A Framework to Enable Two-Layer Inference for Ambient Intelligence
7
GPS packet with user’s name, longitude and latitude data is like this in EN format: [urn:uuid:7bcf39 "JohnSmith" "25.47" "65.06"]
This packet can also be transformed into a RDF statement. In this scenario, high level inference can be applied for interaction among multiple devices. For example, mobile map applications on the device can estimate the driving distance of the parents. And high level inference can decide who should visit the kindergarten by using a following SWRL rule: Implies(Antecedent drivingdistance(?person1, ?dis1) drivingdistance(?person2, ?dis2) swrlb:greaterThan(?dis1, ?dis2) Consequent(showReminder(?person1 "Please visit kindergarten to pick up your kids") showReminder(?person2 "Please go home directly")) )
The above example shows the utilization of the inference engine at two levels, which enable reasoning of different functionalities. More complex reasoning can be performed based on the data from more sensors and users in a real implementation.
5 Discussion We presented a novel framework to enable two-layer inference in ambient environments. This framework can utilize the computation capability of the system at its best, while minimizing the communications. We introduced the requirements, technologies, and design considerations in details. A use case has been presented to illustrate the usefulness of our framework. Three kinds of devices with different capabilities had been considered: resourceconstrained sensors, ontology-enabled mobile terminals and servers. Two-layer inference framework can be deployed on ontology-enabled terminals and servers respectively. Ontology-enabled terminals support fast and local reasoning to react the changing environment, while servers afford complex inference and other resourceconsuming semantic functionalities. The proposed framework is a scalable solution as the utilization of unique identifiers (Universally Unique Identifiers and URIs) in EN packets. As future work, we will verify this framework with a real world implementation, which includes the server, mobile devices and very small sensors that can send measurements in the EN format. Moreover, we will study the structure of the domain ontology to facilitate the sharing of knowledge in a plug-in fashion. Finally, we will consider the optimization of the inference engine to minimize the computing resources for mobile devices.
8
Xiang Su, Davide Fucci and Jukka Riekki
Acknowledgements We thank Dr. Jouni Markkula and Ekaterina Gilman for their advice for this paper. The first author would thank the funding from the Infotech Oulu Graduate School and the MOTIVE research program of the Academy of Finland.
References 1. Alessandra Agostini, Claudio Bettini, Daniele Riboni (2005) Loosely Coupling Ontological Reasoning with an Efficient Middleware for Context-awareness. In: Proceedings of the 2nd Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous 2005), pp.175-182 2. Chang Xu, Shing-Chi Cheung, Cindy Lo et al (2004) Cabot: On the Ontology for the Middleware Support of Context-Aware Pervasive Applications. In: Network and Parallel Computing, pp. 568-575 3. Dave Beckett (2004) RDF/XML Syntax Specification (Revised). The World Wide Web Consortium Recommendation. http://www.w3.org/TR/2004/REC-rdf-syntax-grammar20040210/. 4. Deborah L. McGuinness, Frank van Harmelen (2004) OWL: Web Ontology Language Overview. The World Wide Web Consortium Recommendation. http://www.w3.org/TR/2004/REC-owl-features-20040210/. 5. E. Goh, D. Chieng, A. K. Mustapha et al (2007) A Context-Aware Architecture for Smart Space Environment. In: Multimedia and Ubiquitous Engineering, pp. 908-913 6. Giovanni Acampora, Vincenzo Loia, Michele Nappi et al (2005) Hybrid Computational Intelligence for Ambient Intelligent Environments. In: Atlantic Web Intelligence Conference, pp. 26-31 7. Ian Horrocks, Peter F. Patel-Schneider, Harold Boley et al (2004) SWRL: A Semantic Web Rule Language Combining OWL and RuleML. The World Wide Web Consortium Recommendation. http://www.w3.org/Submission/2004/SUBM-SWRL-20040521/. 8. Ivan Herman (2006) Overview of the W3C Technologies. The World Wide Web Consortium. http://www.w3.org/Consortium/Offices/Presentations/Overview/(6). 9. Jeremy J. Carroll, Graham Klyne (2004) Resource Description Framework (RDF): Concepts and Abstract Syntax. The World Wide Web Consortium Recommendation. http://www.w3.org/TR/2004/REC-rdf-concepts-20040210/. 10. Riekki Jukka., Su Xiang., Janne Haverinen (2008) Connecting Resource-constrained Robots to Knowledge-based Systems. In: Proceedings of International Conference on Modelling, Identification, and Control (MIC 2008) 11. Wang, K.I., Abdulla, W.H., Salcic, Z. (2006) Multi-agent fuzzy inference control system for intelligent environments using Jade. In: 2nd IET International Conference on Intelligent Environments (IE 06), pp. 285-294