It can be argued that the future of machines lies in embedding intelligent systems within them. Unfortunately, sophisticated artificial intelligence (AI) software is usually large and complicated, requiring powerful processors that are not practical in most embedded systems owing to cost, size, and heat production. One solution is to distribute the intelligent processing across many less powerful processors. This research has investigated the suitability, characteristics, and potential of a distributed blackboard system as an· architecture for the implementation of complex AI software in an embedded distributed processing network. A distributed blackboard system called DARBS (Distributed Algorithmic and Rule-based Blackboard System) has been implemented on a distributed processing network comprising up to 18 personal computers. Using the TileWorld environment as a test-bed, the distributed set-up was found to outperform a non-distributed one, although both were non-optimal. The speedup factor of the distributed blackboard system increases up to a maximum as the number of agent processors (APs) increases, after which it drops a little and levels off. To obtain maximum speedup for a given number of agents, it was found that an even distribution of agents across the APs is required, while avoiding saturation of the blackboard. The optimum number of agents per AP was found to be two, with blackboard saturation starting at eight APs. Based on these findings, an embedded version of DARBS (called emDARBS) was designed and implemented on a SARNet parallel processing network of low-cost StrongARM processors. This implementation has demonstrated that a distributed blackboard system is suitable for embedded distributed processing networks, but that some changes are required to tailor it for embedded systems. The potential for embedding intelligent systems in everyday machines has thus been demonstrated.

Real world environments are characterized by high levels of linguistic and numerical uncertainties [Sahab 2011 c]. A Fuzzy Logic System' (FLS) is recognized as an adequate methodology to handle the uncertainties and imprecision available in real world environments and applications. Since the invention of fuzzy logic, it has been applied with great success to numerous real world applications. The first generation of FLSs was type-1 FLSs in which type-1 fuzzy sets were employed. Later, it was found that using type-2 FLSs can enable the handling of higher levels of uncertainties. Recent works have shown that interval type-2 FLSs can outperform type-1 FLSs in the applications which encompass high uncertainty levels. However, the majority of interval type-2 FLSs handle the linguistic and input numerical uncertainties using singleton interval type-2 FLSs that mix the numerical and linguistic uncertainties to be handled only by the linguistic labels type-2 fuzzy sets. This ignores the fact that if input numerical uncertainties were present, they should affect the inputs to the FLS. Even in the papers that employed non-singleton type-2 FLSs, the input signals were assumed to have a predefined shape (mostly Gaussian or triangular) which might not reflect the real uncertainty distribution which can vary with the associated measurement. In this thesis, we will present a new approach which is based on an adaptive non- singleton interval type-2 FLS where the numerical uncertainties will be modeled and handled by non-singleton type-2 fuzzy inputs and the linguistic uncertainties will be handled by interval type-2 fuzzy sets to represent the antecedents' linguistic labels. The non-singleton type-2 fuzzy inputs are dynamic and they are automatically generated from data and they do not assume a specific shape about the distribution associated with the given sensor.

This thesis explores using Dooyeweerdian philosophy as a foundation for constructing knowledge-based system (KBS) in complex, ill-structured and human intensive domains. Dooyeweerd's transcendental philosophy is a relatively recent phenomenon in the medium of the English language, and has only recently been considered in relation to knowledge-based systems and sustainability. Consequently, this thesis has little background research to draw upon and serves as a prolegomena - a 'first-word' on the subject matter.

This thesis explores the reuse of knowledge bases through semi-automated code generation of new KBs and explores the issues surrounding the use of constraint satisfaction techniques to solve the Sisyphus II VT challenge and the related class of problems.  For my primary analysis I chose a knowledge base system written in CLIPS which was based on the propose-and-revise (PnR) problem solver, and which had a lift/elevator knowledge base (KB).  Components were initially extracted manually, and then semi-automatically, and were used with both an Excel spreadsheet and a constraint problem solver (ECLiPSe) to solve a range of tasks.  I have produced a reuse system which is useable by domain experts. The next phase was to implement ExtrAKTor which extracts the same 4 knowledge sources virtually automatically from the CLIPS knowledge base (held by Protégé), and to transform these so that they are compatible with additional problem solvers.  To date Excel & ECLiPSe have been used, and again it has been demonstrated that the resulting systems are able to solve lift configuration tasks.  This work has produced a reuse system which is useable by domain experts.  The ExtrAKTor system has been independently tested using an enhanced version of the U-HAUL KB; a KB in the removal truck rental domain. Fixes are an essential part of P&P.  Using the tightly constrained VT KB generated by ExtrAKTor from the original Protégé KB, ECLiPSe did not require fix information to find a solution.  The decision was made to create a less tightly constrained version of the KB to more fully investigate the solution space and determine if fix information could be encoded and used effectively in a CSP environment.  It became apparent that performance was significantly affected by ordering of clauses due to Prolog’s backtracking during component selection; this had not been an issue with the more tightly constrained KB.  The ExtrAKTor code was then further refined adding the “domain” and “infers most” construct which led to a substantial performance improvement largely independent of clause ordering.  The conclusion of the experimentation is that ECLiPSe does not require expert provided fix information to efficiently solve the VT class of parametric design problems.

No description available.

No description available.

This thesis aims to explore and demonstrate the ability of the new standard of structural and behavioural components in Unified Modelling Language (UML 2.0 / 2004) to model the learning behaviour of Intelligent Agents. The thesis adopts the research direction that views agent-oriented systems as an extension to object-oriented systems. In view of the fact that UML has been the de facto standard for modelling object-oriented systems, this thesis concentrates on exploring such modelling potential with Intelligent Agent-oriented systems. Intelligent Agents are Agents that have the capability to learn and reach agreement with other Agents or users. The research focuses on modelling the learning behaviour of a single Intelligent Agent, as it is the core of multi-agent systems. During the writing of the thesis, the only work done to use UML 2.0 to model structural components of Agents was from the Foundation for Intelligent Physical Agent (FIPA). The research builds upon, explores, and utilises this work and provides further development to model the structural components of learning behaviour of Intelligent Agents. The research also shows the ability of UML version 2.0 behaviour diagrams, namely activity diagrams and sequence diagrams, to model the learning behaviour of Intelligent Agents that use learning from observation and discovery as well as learning from examples of strategies. The research also evaluates if UML 2.0 state machine diagrams can model specific reinforcement learning algorithms, namely dynamic programming, Monte Carlo, and temporal difference algorithms. The thesis includes user guides of UML 2.0 activity, sequence, and state machine diagrams to allow researchers in agent-oriented systems to use the UML 2.0 diagrams in modelling the learning components of Intelligent Agents. The capacity for learning is a crucial feature of Intelligent Agents. The research identifies different learning components required to model the learning behaviour of Intelligent Agents such as learning goals, learning strategies, and learning feedback methods. In recent years, the Agent-oriented research has been geared towards the agency dimension of Intelligent Agents. Thus, there is a need to conduct more research on the intelligence dimension of Intelligent Agents, such as negotiation and argumentation skills. The research shows that behavioural components of UML 2.0 are capable of modelling the learning behaviour of Intelligent Agents while structural components of UML 2.0 need extension to cover structural requirements of Agents and Intelligent Agents. UML 2.0 has an extension mechanism to fulfil Agents and Intelligent Agents for such requirements. This thesis will lead to increasing interest in the intelligence dimension rather than the agency dimension of Intelligent Agents, and pave the way for objectoriented methodologies to shift more easily to paradigms of Intelligent Agent-oriented systems.

Recommender systems have been widely advocated as a way of coping with the problem of information overload for knowledge workers. Given this, multiple recommendation methods have been developed. However, it has been shown that no one technique is best for all users in all situations. Thus, we believe that effective recommender systems should incorporate a wide variety of such techniques and that some form of overarching framework should be put in place to coordinate the various recommendations so that only the best of them (from whatever source) are presented to the user. To this end, we show that a marketplace, in which the various recommendation methods compete to offer their recommendations to the user, can be used in this role. Specifically, our research is concerned with the principled design of such a marketplace (including the auction protocol, the reward mechanism and the bidding strategies of the individual recommender agents) and its evaluation in terms of how it can effectively coordinate multiple methods. In addition to the market mechanisms, a reinforcement learning strategy is developed to assist the individual recommender agents' bidding behaviour so as to learn the users' interests and still maximize their revenue. Finally, we evaluate our approach with a real market-based recommender system that is composed of a number of typical recommendation methods and that is evaluated with real users. The evaluation results show that our approach is indeed an effective means of coordinating multiple different recommendation methods in one single system and is an effective way of dealing with the problem of information overload.

The main contribution of the thesis is to present a systematic process to develop an agent-based system that assists a system developer to construct the required system, through a series of modelling activities, employing several levels of abstraction to show the milestones and produce intermediate deliverables. Current practice emphasises "downstream activities" such as implementation at the expense of "upstream activities" such as "modelling". The research has found that the development process for an agent system consists of three phases: agent system development, agent environment development and agent system deployment. The first and second phases represent an intertwined spiral model. All three phases themselves consist of three stages. Each phase employs different development techniques and each stage uses appropriate models and tools such as problem domain model, agent use cases, scenarios, agent system architecture, plan model and individual agent model. The proposed agent development method is applied to two case studies: a Filtering Agent System and Diabetic Consultation System. Both systems have been implemented and tested. Three distinct ways were used to evaluate the proposed method. First, comparing with the criteria of a methodology. Second, comparing it with the current agent-oriented methodologies. Third, informal observations from a potential user community. In conclusion, the research has demonstrated an effective synthesising process to build a set of agent concepts, development life-cycle and modelling to show a systematic process for developing agent systems. Moreover, by employing a whole host of software abstraction tools and techniques in the process, two benefits accrue: the introduction of more 'up stream' activities as well as placing modelling at the heart of the process. Illustratively, we could say that the modelling presented here does for agent systems what data flow diagram and data entity diagram have done for structured methodologies, i.e. raise the level of abstraction employed.

A large range of products are now designed based on the implementation of microcontrollers, as they lead to an overall system enhancement, such as a more compact system design and reduced costs. This can be credited to the peripheral functions embodied in the microcontroller and the flexibility offered by its software. Since the applications of microcontrollers are achieved through software, the implementation of the target controller is considered as an interdisciplinary process. Due to the capability of the current development tool, designers are required to input significant effort such that the development with the process is taking a long time to complete. The purpose of this research is aimed at improving the current microcontroller software development process.