Search

A life expectancy beyond 60 years old provides us with unprecedented life opportunities. However, it also increases our risk of suffering from age-related health conditions that would prevent us from enjoying such opportunities. Dementia is one of the main such conditions. The most frequent cause of dementia is Alzheimer's Disease (AD), which is known to start silently up to 20 years before observable symptoms appear. Therefore, interdisciplinary research efforts are focusing on unravelling disease progression and developing strategies for secondary prevention (i.e. where disease is present, but no symptoms have emerged). Whilst memory loss is usually regarded as the most prominent of these symptoms, language decline is also an essential source of clinical information that is gradually receiving more attention, given language ubiquity. The development of artificial intelligence (AI), particularly speech analysis, natural language processing and machine learning, offers opportunities for the automatic analysis of spoken language data. The work presented in this thesis applies an AI approach to the context of dementia, addressing three interrelated research questions, [1] is automatic speech analysis a viable approach to detecting pre-clinical stages of dementia? [2] if so, what type of language (i.e. monologue vs. dialogue, conversational vs. narrative, spontaneous vs. scripted) is most suitable for such analysis? and [3] which characteristics of spoken language are most predictive and practically useful (i.e. linguistic vs. acoustic)? In order to answer these questions, a systematic review of the last twenty years of literature is conducted to assess the state-of-the-art. The results of such a review indicate that this is a very promising emerging field, although research outcomes are still far from being translated into clinical practice and medical research. Subsequently, a computational analysis is performed in order to compare older adults with and without dementia on the basis of features extracted from their spoken language. The two main datasets which are available in the field are used. One of them is a monologue, task-based corpus: the Pitt Corpus; and the other one is a dialogue, conversation-based corpus: the Carolina Conversations Collection (CCC). Acoustic features are extracted from both datasets and used as input for machine learning classification, achieving a classification accuracy of up to 78.70% for Pitt Corpus, and up to 85.21% for CCC. The extracted acoustic features consist of a comprehensive set of prosodic parameters. Finally, the Prevent-ED project is presented, which was specifically designed in the context of this thesis in order to address some of the limitations gathered from the review and experiments. Prevent-ED is a protocol for a novel study design, where spontaneous language is recorded by prompting natural and rich conversations over the map of an imaginary land. Different to previous research, which involved already diagnosed cohorts, this project involves middle-age healthy adults at different levels of risk of developing dementia later in life. Accordingly, the main objective is to investigate whether speech features are predictive of such levels of risk, instead of diagnostic labels. Therefore, the design, development, implementation and dissemination of the Prevent-ED project constitute a major contribution of this doctoral research. Moreover, new data has been collected through this project and preliminary analyses show a promising tendency that is likely to be strengthened when the data collection is complete, currently on hold due to COVID-19. In conclusion, the contribution of this doctoral research is three-fold: an extensive systematic review, the first one of the field; an analysis of the two main available datasets containing speech produced by individuals with and without dementia; and Prevent-ED, a novel protocol for data collection and analysis, which attempts to offer further insights into the "at risk" stages of the disease, different to previous research on already diagnosed cohorts. The resulting high-quality database will be distributed to the wider research community.

Embryonic stem cells (ESCs) hold great potential for developing future therapies for a wide range of diseases. However, the mechanisms of pattern formation during embryonic development remain poorly understood. ESCs in culture self-organise to form spatial patterns of gene expression upon geometrical confinement indicating that patterning is an emergent phenomenon that results from the many interactions between the cells. Here, we applied an agent-based modelling approach to identify biologically plausible rules acting at the mesoscale within stem cell collectives that may explain spontaneous patterning. We tested different models involving differential motile behaviours including exploring effects due to neighbour interactions. We introduced a new metric, the stem cell aggregate pattern distance (SCAPD), to assess the deviation between the probabilistic experimental pattern formation (used as ground truth) and the probabilistic simulated outcome. We demonstrated our models can produce broadly realistic pattern formation (when compared to experimental data) with a quantified level of uncertainty. The best of our models improve fitness, evaluated by SCAPD, by 70% and 77% over the random models for a discoidal or an ellipsoidal stem cell confinement, respectively. Collectively, our findings provide compelling arguments that a parsimonious mechanism that involves differential motility is sufficient to explain the spontaneous patterning of the cells upon confinement. Furthermore, our work also defines a region of the parameter space that is compatible with patterning, which assists future studies in the field of cell engineering. We envisage that the novel approaches explored within this work will be applicable to many biological systems and will contribute towards facilitating progress by reducing the need for extensive and costly experiments.

There is an emerging class of applications that require long-term tracking and monitoring of mobile entities for characterising their contexts and behaviours using data from wireless sensors. Examples include monitoring animals in their natural habitat over the annual cycle; tracking shipping containers and their handling during transit; and monitoring air quality using sensors attached to bicycles used in public sharing schemes. All applications within this class require the acquisition of sensor data tagged with spatio-temporal information and uploaded wirelessly. Currently there is no solution targeting the entire class of applications, only point solutions focused on specific scenarios. This thesis presents a complete solution (firmware and hardware) for applications within this class that consists of attaching mobile sensor nodes to the entities for tracking and monitoring their behaviour, and deploying an infrastructure of base-stations for collecting the data wirelessly. The proposed solution is more energy efficient compared to the existing solutions that target specific scenarios, offering a longer deployment lifetime with a reduced size and weight of the devices. This is achieved mainly by using the VB-TDMA low-power data upload protocol proposed in this thesis. The mobile sensor nodes, consisting of the GPS and radio modules among others, and the base-stations are powered by batteries, and the optimisation of their energy usage is of primary concern. The presence of the GPS module, in particular its acquisition of accurate time, is used by the VB-TDMA protocol to synchronise the communication between nodes at no additional energy costs, resulting in an energy-efficient data upload protocol for sparse networks of mobile nodes, that can potentially be out of range of base-stations for extended periods of time. The VB-TDMA and an asynchronous data upload protocol were implemented on the custom-designed Prospeckz-5-based wireless sensor nodes. The protocols’ performances were simulated in the SpeckSim simulator and validated in real-world deployments of tracking and monitoring thirty-two Retuerta wild horses in the Doñana National Park in Spain, and a herd of domesticated horses in Edinburgh. The chosen test scenario of long-term wildlife tracking and monitoring is representative for the targeted class of applications. The VB-TDMA protocol showed a significantly lower power consumption than other comparable MAC protocols, effectively doubling the battery lifetime. The main contributions of the thesis are the development of the VB-TDMA data upload protocol and its performance evaluation, along with the development of simulation models for performance analysis of wireless sensor networks, validated using data from the two real-world deployments.

Self presentation is evolving; with digital technologies, with the Web and personal publishing, and then with mainstream adoption of online social media. Where are we going next? One possibility is towards a world where we log and own vast amounts of data about ourselves. We choose to share - or not - the data as part of our identity, and in interactions with others; it contributes to our day-to-day personhood or sense of self. I imagine a world where the individual is empowered by their digital traces (not imprisoned), but this is a complex world. This thesis examines the many factors at play when we present ourselves through Web technologies. I optimistically look to a future where control over our digital identities are not in the hands of centralised actors, but our own, and both survey and contribute to the ongoing technical work which strives to make this a reality. Decentralisation changes things in unexpected ways. In the context of the bigger picture of our online selves, building on what we already know about self-presentation from decades of Social Science research, I examine what might change as we move towards decentralisation; how people could be affected, and what the possibilities are for a positive change. Finally I explore one possible way of self-presentation on a decentralised social Web through lightweight controls which allow an audience to set their expectations in order for the subject to meet them appropriately. I seek to acknowledge the multifaceted, complicated, messy, socially-shaped nature of the self in a way that makes sense to software developers. Technology may always fall short when dealing with humanness, but the framework outlined in this thesis can provide a foundation for more easily considering all of the factors surrounding individual self-presentation in order to build future systems which empower participants.

Scientific data does not exist on the Web in the same way as the written word; reviews, media, wikis, social networks, and blogs all contribute to the interconnected nature of ordinary language on the Web. Network effects create additional value from seemingly minor contributions to the Web. But nothing such as this exists for scientific data. Simply put, within the Open Web Platform, we cannot currently turn and apply similar mechanisms for scientific work without great effort. Thus, the Web has not so far enabled Science as well as it has enabled dissemination and interconnection for the written word: to truly enable Science on the Web, we must endeavor to make data and its semantics first-class Web constituents. This thesis focuses on solving this problem by enabling scientific data to exist on the Web in such a way that it can be processed both as viewable content and consumed data. Starting from the principles on which the Web has so far thrived, we propose solutions to enable complex data exchanges while preserving the Web as it stands. We introduce the Partition Annotate Name (PAN) methodology, which relies upon embracing the core architectural principles of the Web: name things with URIs; process common data formats; use common rules under a shared contract between publisher, developer, and consumer.

One way to build large-scale autonomous systems is to develop an open multi-agent system using peer-to-peer architectures in which agents are not pre-engineered to work together and in which agents themselves determine the social norms that govern collective behaviour. The social norms and the agent interaction models can be described by Electronic Institutions such as those expressed in the Lightweight Coordination Calculus (LCC), a compact executable specification language based on logic programming and pi-calculus. Open multi-agent systems have experienced growing popularity in the multi-agent community and are expected to have many applications in the near future as large scale distributed systems become more widespread, e.g. in emergency response, electronic commerce and cloud computing. A major practical limitation to such systems is security, because the very openness of such systems opens the doors to adversaries for exploit existing vulnerabilities. This thesis addresses the security of open multi-agent systems governed by electronic institutions. First, the main forms of attack on open multi-agent systems are introduced and classified in the proposed attack taxonomy. Then, various security techniques from the literature are surveyed and analysed. These techniques are categorised as either prevention or detection approaches. Appropriate countermeasures to each class of attack are also suggested. A fundamental limitation of conventional security mechanisms (e.g. access control and encryption) is the inability to prevent information from being propagated. Focusing on information leakage in choreography systems using LCC, we then suggest two frameworks to detect insecure information flows: conceptual modeling of interaction models and language-based information flow analysis. A novel security-typed LCC language is proposed to address the latter approach. Both static (design-time) and dynamic (run-time) security type checking are employed to guarantee no information leakage can occur in annotated LCC interaction models. The proposed security type system is then formally evaluated by proving its properties. A limitation of both conceptual modeling and language-based frameworks is difficulty of formalising realistic policies using annotations. Finally, the proposed security-typed LCC is applied to a cloud computing configuration case study, in which virtual machine migration is managed. The secrecy of LCC interaction models for virtual machine management is analysed and information leaks are discussed.

The Internet and Web have brought in a new era of information sharing and opened up countless opportunities for people to rethink and redefine communication. With the development of network-related technologies, a Client/Server architecture has become dominant in the application layer of the Internet. Nowadays network nodes are behind firewalls and Network Address Translations, and the centralised design of the Client/Server architecture limits communication between users on the client side. Achieving the conflicting goals of data privacy and data openness is difficult and in many cases the difficulty is compounded by the differing solutions adopted by different organisations and companies. Building a more decentralised or distributed environment for people to freely share their knowledge has become a pressing challenge and we need to understand how to adapt the pervasive Client/Server architecture to this more fluid environment. This thesis describes a novel framework by which network nodes or humans can interact and share knowledge with each other through formal service-choreography specifications in a decentralised manner. The platform allows peers to publish, discover and (un)subscribe to those specifications in the form of Interaction Models (IMs). Peer groups can be dynamically formed and disbanded based on the interaction logs of peers. IMs are published in HTML documents as normal Web pages indexable by search engines and associated with lightweight annotations which semantically enhance the embedded IM elements and at the same time make IM publications comply with the Linked Data principles. The execution of IMs is decentralised on each peer via conventional Web browsers, potentially giving the system access to a very large user community. In this thesis, after developing a proof-of-concept implementation, we carry out case studies of the resulting functionality and evaluate the implementation across several metrics. An increasing number of service providers have began to look for customers proactively, and we believe that in the near future we will not search for services but rather services will find us through our peer communities. Our approaches show how a peer-to-peer architecture for this purpose can be obtained on top of a conventional Client/Server Web infrastructure.

As multi-agent systems (MAS) have become more mature and systems in general have become more distributed, it is necessary for those who want to build large scale systems to consider, in some computational depth, how agents can communicate in large scale, complex and distributed systems. Currently, some MAS systems have been developed to use an abstract specification language for argumentation. This as a basis for agent communication; to provide effective decision support for agents and yield better agreements. However, as we build complete MAS that involve argumentation, there is a need to produce concrete implementations in which these abstract specifications are realised via protocols coordinating agent behaviour. This creates a gap between standard argument specification and deployment of protocols. This thesis attempts to close this gap by using a combination of automated synthesis and verification methods. More precisely, this thesis proposes a means of moving rapidly from argument specification to protocol implementation using an extension of the Argument Interchange Format (AIF is a generic specification language for argument structure) called a Dialogue Interaction Diagram (DID) as the dialogue game specification language and the Lightweight Coordination Calculus (LCC is an executable specification language used for coordinating agents in open systems) as an implementation language. The main contribution of this research is to provide approaches for enabling developers of dialogue game argumentation systems to use specification languages (in our case AIF/DID) to generate agent protocol systems that are capable of direct implementation on open infrastructures (in our case LCC).

Little attention has been paid, in depth, to the relationship between fitness evaluation in evolutionary algorithms and reputation mechanisms in multi-agent systems, but if these could be related it opens the way for implementation of distributed evolutionary systems via multi-agent architectures. Our investigation concentrates on the effectiveness with which social selection, in the form of reputation, can replace direct fitness observation as the selection bias in an evolutionary multi-agent system. We do this in two stages: In the first, we implement a peer-to-peer, adaptive Genetic Algorithm (GA), in which agents act as individual GAs that, in turn, evolve dynamically themselves in real-time, using the traditional evolutionary operators of fitness-based selection, crossover and mutation. In the second stage, we replace the fitness-based selection operator with a reputation-based one, in which agents choose their mates based on the collective past experiences of themselves and their peers. Our investigation shows that this simple model of distributed reputation can be successful as the evolutionary drive in such a system, exhibiting practically identical performance and scalability to direct fitness observation. Further, we discuss the effect of noise (in the form of “defective” agents) in both models. We show that the reputation-based model is significantly better at identifying the defective agents, thus showing an increased level of resistance to noise.

Ensemble learning is a promising direction of research in machine learning, in which an ensemble classifier gives better predictive and more robust performance for classification problems by combining other learners. Meanwhile agent-based systems provide frameworks to share knowledge from multiple agents in an open context. This thesis combines multi-agent knowledge sharing with ensemble methods to produce a new style of learning system for open environments. We now are surrounded by many smart objects such as wireless sensors, ambient communication devices, mobile medical devices and even information supplied via other humans. When we coordinate smart objects properly, we can produce a form of collective intelligence from their collaboration. Traditional ensemble methods and agent-based systems have complementary advantages and disadvantages in this context. Traditional ensemble methods show better classification performance, while agent-based systems might not guarantee their performance for classification. Traditional ensemble methods work as closed and centralised systems (so they cannot handle classifiers in an open context), while agent-based systems are natural vehicles for classifiers in an open context. We designed an open and social ensemble learning architecture, named J-model, to merge the conflicting benefits of the two research domains. The J-model architecture is based on a service choreography approach for coordinating classifiers. Coordination protocols are defined by interaction models that describe how classifiers will interact with one another in a peer-to-peer manner. The peer ranking algorithm recommends more appropriate classifiers to participate in an interaction model to boost the success rate of results of their interactions. Coordinated participant classifiers who are recommended by the peer ranking algorithm become an ensemble classifier within J-model. We evaluated J-model’s classification performance with 13 UCI machine learning benchmark data sets and a virtual screening problem as a realistic classification problem. J-model showed better performance of accuracy, for 9 benchmark sets out of 13 data sets, than 8 other representative traditional ensemble methods. J-model gave better results of specificity for 7 benchmark sets. In the virtual screening problem, J-model gave better results for 12 out of 16 bioassays than already published results. We defined different interaction models for each specific classification task and the peer ranking algorithm was used across all the interaction models. Our research contributions to knowledge are as follows. First, we showed that service choreography can be an effective ensemble coordination method for classifiers in an open context. Second, we used interaction models that implement task specific coordinations of classifiers to solve a variety of representative classification problems. Third, we designed the peer ranking algorithm which is generally and independently applicable to the task of recommending appropriate member classifiers from a classifier pool based on an open pool of interaction models and classifiers.

In this thesis an architecture, similar to subsumption architectures, is presented which uses low level behaviour modules, based on combinations of machine learning techniques, to create teams of autonomous agents cooperating via shared plans for interaction. The purpose of this is to perform effective single plan execution within multiple scenarios, using a modern team based first person shooter video game as the domain and visualiser. The main focus is showing that through basic machine learning mechanisms, applied in a multi-agent setting on sparse data, plans can be executed on game levels of varying size and shape without sacrificing team goals. It is also shown how different team members can perform locally sub-optimal operations which contribute to a globally better strategy by adding exploration data to the machine learning mechanisms. This contributes to the reinforcement learning problem of exploration versus exploitation, from a multi-agent perspective.

Software agents interact to solve tasks, the details of which need to be described in a language understandable by all the actors involved. Ontologies provide a formalism for defining both the domain of the task and the terminology used to describe it. However, finding a shared ontology has proved difficult: different institutions and developers have different needs and formalise them in different ontologies. In a closed environment it is possible to force all the participants to share the same ontology, while in open and distributed environments ontology mapping can provide interoperability between heterogeneous interacting actors. However, conventional mapping systems focus on acquiring static information, and on mapping whole ontologies, which is infeasible in open systems. This thesis shows a different approach to the problem of heterogeneity. It starts from the intuitive idea that when similar situations arise, similar interactions are performed. If the interactions between actors are specified in formal scripts, shared by all the participants, then when the same situation arises, the same script is used. The main hypothesis that this thesis aims to demonstrate is that by analysing different runs of these scripts it is possible to create a statistical model of the interactions, that reflect the frequency of terms in messages and of ontological relations between terms in different messages. The model is then used during a run of a known interaction to compute the probability distribution for terms in received messages. The probability distribution provides additional information, contextual to the interaction, that can be used by a traditional ontology matcher in order to improve efficiency, by reducing the comparisons to the most likely ones given the context, and possibly both recall and precision, in particular helping disambiguation. The ability to create a model that reflects real phenomena in this sort of environment is evaluated by analysing the quality of the predictions, in particular verifying how various features of the interactions, such as their non-stationarity, affect the predictions. The actual improvements to a matcher we developed are also evaluated. The overall results are very promising, as using the predictor can lower the overall computation time for matching by ten times, while maintaining or in some cases improving recall and precision.

In distributed open systems, such as multiagent systems, new interactions are constantly appearing and new agents are continuously joining or leaving. It is unrealistic to expect agents to automatically trust new interactions. It is also unrealistic to expect agents to refer to their users for help every time a new interaction is encountered. An agent should decide for itself whether a specific interaction with a given group of agents is suitable or not. This thesis presents a runtime verification mechanism for addressing this problem. Verifying multiagent systems has its challenges. It is hard to predict the reliability of interactions, in systems that are heavily influenced by autonomous agents, without having access to the agent specifications. Available verification mechanisms may roughly be divided into two categories: (1) those that verify interaction models independently of specific agents, and (2) those that verify agent models whose constraints shape the interactions. Interaction models are not sufficient when verifying dynamic properties that depend on the agents engaged in an interaction. On the other hand, verifying agent specifications, such as BDI models, is extremely inefficient. Specifications are usually not explicit enough, resulting in the verification of a massive number of permissible interactions. Furthermore, in open systems, an agent’s internal specification is usually not accessible for many reasons, including security and privacy. This thesis proposes a model checker that verifies a combination of a global interaction model and local deontic models. The deontic model may be viewed as a list of agent constraints that are deemed necessary to share and verify, such as the inability of the buyer to pay by credit card. The result is a lightweight, efficient, and powerful model checker that is capable of verifying rich properties of multiagent systems without the need for accessing agents’ internal specifications. Although the proposed model checker has potential for addressing a variety of problems, the trust domain receives special attention due to the critically of the trust issue in distributed open systems and the lack of reliable trust solutions. The thesis illustrates how a dynamic model checker, using deontic/trust models, can help agents decide whether the scenarios they wish to join are trustworthy or not. In summary, the main contribution of this research is in introducing interaction time verification for checking deontic and trust models multiagent interactions. When faced with new unexplored interactions, agents can verify whether joining a given interaction with a given set of collaborating agents would violate any of its constraints.

Service-oriented architectures are a popular architectural paradigm for building software applications from a number of loosely coupled, distributed services. Through a set of procedural rules, workflow technologies define how groups of services coordinate with one another to achieve a shared task. A problem with workflow specifications is that often the patterns of interaction between the distributed services are too complicated to predict and analyse at design-time. In certain cases, the exact patterns of message exchange and the concrete services to call cannot be predicted in advance, due to factors such as fluctuating network load or the availability of services. It is a more realistic assumption to endow software components with the ability to make decisions about the nature and scope of their interactions at runtime. Multiagent systems offer a complementary paradigm: building software applications from a number of self interested, autonomous agents. This thesis presents an investigation into fusing the agency and service-oriented architecture paradigms, in order to facilitate flexible, workflow composition. Our approach offers an agent-based solution to service choreography and is founded on the concept of shared interaction protocols. By adopting an agent-based approach to service choreography, active autonomous agents can utilise the typically passive service-oriented architectures, found in Internet and Grid systems. In contrast with statically defined, centralised service orchestrations, decentralised agents can perform service choreography at runtime, allowing them to operate in scenarios where it is not possible to define the pattern of interaction in advance. Application to real scenarios is a driving factor behind this research. By working closely with a number of active Grid projects, namely AstroGrid and the Large-Synoptic Survey Telescope (LSST), a concrete set of requirements for scientific workflow have been derived, based on realistic science problems. This research has resulted in the MultiAgent Service Choreography (MASC) language to express scientific workflow, methodology for system building and a software framework which performs agent based Web service choreography, in order to enact distributed e-Science experiments. Evaluation of this thesis is conducted through case study, applying the language, methodology and software framework to solve a motivating set of workflow scenarios.