Many complex systems in nature are governed by simple local interactions, although a number are also described by global interactions. For example, within the field of hydraulics the Navier-Stokes equations describe free-surface water flow, through means of the global preservation of water volume, momentum and energy. However, solving such partial differential equations (PDEs) is computationally expensive when applied to large 2D flow problems. An alternative which reduces the computational complexity, is to use a local derivative to approximate the PDEs, such as finite difference methods, or Cellular Automata (CA). The high speed processing of such simulations is important to modern scientific investigation especially within urban flood modelling, as urban expansion continues to increase the number of impervious areas that need to be modelled. Large numbers of model runs or large spatial or temporal resolution simulations are required in order to investigate, for example, climate change, early warning systems, and sewer design optimisation. The recent introduction of the Graphics Processor Unit (GPU) as a general purpose computing device (General Purpose Graphical Processor Unit, GPGPU) allows this hardware to be used for the accelerated processing of such locally driven simulations. A novel CA transformation for use with GPUs is proposed here to make maximum use of the GPU hardware. CA models are defined by the local state transition rules, which are used in every cell in parallel, and provide an excellent platform for a comparative study of possible alternative state transition rules. Writing local state transition rules for CA systems is a difficult task for humans due to the number and complexity of possible interactions, and is known as the ‘inverse problem’ for CA. Therefore, the use of Genetic Programming (GP) algorithms for the automatic development of state transition rules from example data is also investigated in this thesis. GP is investigated as it is capable of searching the intractably large areas of possible state transition rules, and producing near optimal solutions. However, such population-based optimisation algorithms are limited by the cost of many repeated evaluations of the fitness function, which in this case requires the comparison of a CA simulation to given target data. Therefore, the use of GPGPU hardware for the accelerated learning of local rules is also developed. Speed-up factors of up to 50 times over serial Central Processing Unit (CPU) processing are achieved on simple CA, up to 5-10 times speedup over the fully parallel CPU for the learning of urban flood modelling rules. Furthermore, it is shown GP can generate rules which perform competitively when compared with human formulated rules. This is achieved with generalisation to unseen terrains using similar input conditions and different spatial/temporal resolutions in this important application domain.

This thesis is concerned with extending properties of regular word languages to richer structures. We consider intricate properties like the relationship between one-way and two-way temporal logics, minimization of automata, and the ability to effectively characterize logics. We investigate whether these properties can be extended to tree languages or word languages over an infinite alphabet. It is known that linear temporal logic (LTL) is as expressive as first-order logic over finite words [Kam68, GPSS80]. LTL is a unidirectional logic, that can only navigate forwards in a word, hence it is quite surprising that it can capture all of first-order logic. In fact, one of the main ideas of the proof of [GPSS80] is to show that the expressiveness of LTL is not increased if modalities for navigating backwards are added. It is also known that an extension of bidirectional LTL to ordered trees, called Conditional XPath, is first-order complete [Mar04]. We investigate whether the unidirectional fragment of Conditional XPath is also first-order complete. We show that this is not the case. In fact we show that there is a strict hierarchy of expressiveness consisting of languages that are all weaker than first-order logic. Unidirectional Conditional XPath is contained in the lowest level of this hierarchy. In the second part of the thesis we consider data word languages. That is, word languages over an infinite alphabet. We extend the theorem of Myhill and Nerode to a class of automata for data word languages, called deterministic finite memory automata (DMA). We give a characterization of the languages that are accepted by DMA, and also provide an algorithm for minimizing DMA. Finally we extend theorems of Büchi, Schützenberger, McNaughton, and Papert to data word languages. A theorem of Büchi states that a language is regular iff it can be defined in monadic second-order logic. Schützenberger, McNaughton, and Papert have provided an effective characterization of first-order logic, that is, an algorithm for deciding whether a regular language can be defined in first-order logic. We provide a counterpart of Büchi's theorem for data languages. More precisely we define a new logic and we show that it has the same expressiveness as non-deterministic finite memory automata. We then turn to a smaller class of data languages, those that are recognized by algebraic objects called orbit finite data monoids. We define a second new logic and show that it can define precisely the languages accepted by orbit finite data monoids. We provide an effective characterization of a first-order variant of this second logic, as well as of restrictions of first-order logic, such as its two variable fragment and local variants.

Today, there is a substantial number of software and research groups that focus on the development of image processing software to extract useful information from medical images, in order to assist and improve patient diagnosis. The work presented in this thesis is centred on processing of images of blood and bone marrow smears of patients suffering from leukaemia, a common type of cancer. In general, cancer is due to aberrant gene expression, which is caused by either mutations or epigenetic changes in DNA. Poor diet and unhealthy lifestyle may trigger or contribute to these changes, although the underlying mechanism is often unknown. Importantly, many cancer types including leukaemia are curable and patient survival and treatment can be improved, subject to prompt diagnosis. In particular, this study focuses on Acute Myeloid Leukaemia (AML), which can be of eight distinct types (M0 to M7), with the main objective to develop a methodology to automatically detect and classify leukaemia cells into one of the above types. The data was collected from the Department of Haematology, Universiti Sains Malaysia, in Malaysia. Three main methods, namely Cellular Automata, Heuristic Search and classification using Neural Networks are facilitated. In the case of Cellular Automata, an improved method based on the 8-neighbourhood and rules were developed to remove noise from images and estimate the radius of the potential blast cells contained in them. The proposed methodology selects the starting points, corresponding to potential blast cells, for the subsequent seeded heuristic search. The Seeded Heuristic employs a new fitness function for blast cell detection. Furthermore, the WEKA software is utilised for classification of blast cells and hence images, into AML subtypes. As a result accuracy of 97.22% was achieved in the classification of blasts into M3 and other AML subtypes. Finally, these algorithms are integrated into an automated system for image processing. In brief, the research presented in this thesis involves the use of advanced computational techniques for processing and classification of medical images, that is, images of blood samples from patients suffering from leukaemia.

Les automates cellulaires sont un modèle de calcul massivement parallèle introduit dans les années 50. De nombreuses variantes peuvent être considérées par exemple en faisant varier la dimension de l’espace de calcul, ou les possibilités de communication entre les différentes cellules. En effet, chaque cellule ne peut communiquer qu’avec un nombre fini d’autres cellules que l’on appelle son voisinage. Mes travaux s’intéressent principalement à l’impact du choix du voisinage sur les capacités algorithmiques de ce modèle. Cet impact étant bien compris en une dimension, mes travaux portent majoritairement sur les automates cellulaires bidimensionnels. J’ai tout d’abord essayé de généraliser des propriétés classiques de certaines classes de complexité au plus de voisinages possibles. On arrive notamment à un théorème d’accélération linéaire valable pour tous les voisinages. J’ai ensuite étudié les différences entre les classes de faibles complexités en fonction du voisinage choisi. Ces travaux ont permis d’exhiber des voisinages définissant des classes incomparables, ainsi que des ensembles de voisinages définissant exactement les mêmes classes de complexité. Enfin, je présente aussi des travaux sur les différences de puissance de calcul entre les automates de dimensions différentes. / Cellular automata were introduced in the 50s by J. von Neumann and S. Ulamas an efficient way of modeling massively parallel computation. Many variations of the model can be considered such as varying the dimension of the computation space or the communication capabilities of the computing cells. In a cellular automaton each cell can communicate only with a finite number of other cells called its neighbors. My work focuses on the impact of the choice of the neighbors on the algorithmic properties of the model. My first goal was to generalize some classical properties of computation models to the widest possible class of neighborhoods, in particular I prove a linear speedup theorem for any two dimensional neighborhood. I then study the difference between the complexity classes defined by different neighborhoods, show the existence of neighborhoods defining incomparable classes, and some sets of neighborhoods defining identical classes. Finally, I also discuss the impact of the dimension of the automata on their computational power.

Introducing robots into human environments requires them to handle settings designed specifically for human size and morphology, however, large, conventional humanoid robots with stiff, high powered joint actuators pose a significant danger to humans. By contrast, “anthropomimetic” robots mimic both human morphology and internal structure; skeleton, muscles, compliance and high redundancy. Although far safer, their resultant compliant structure presents a formidable challenge to conventional control. Here we review, and seek to address, characteristic control issues of this class of robot, whilst exploiting their biomimetic nature by drawing upon biological motor control research. We derive a novel learning controller for discovering effective reaching actions created through sustained activation of one or more muscle synergies, an approach which draws upon strong, recent evidence from animal and humans studies, but is almost unexplored to date in musculoskeletal robot literature. Since the best synergies for a given robot will be unknown, we derive a deliberately simple reinforcement learning approach intended to allow their emergence, in particular those patterns which aid linearization of control. We also draw upon optimal control theories to encourage the emergence of smoother movement by incorporating signal dependent noise and trial repetition. In addition, we argue the utility of developing a detailed dynamic model of a complete robot and present a stable, physics-based model, of the anthropomimetic ECCERobot, running in real time with 55 muscles and 88 degrees of freedom. Using the model, we find that effective reaching actions can be learned which employ only two sequential motor co-activation patterns, each controlled by just a single common driving signal. Factor analysis shows the emergent muscle co-activations can be reconstructed to significant accuracy using weighted combinations of only 13 common fragments, labelled “candidate synergies”. Using these synergies as drivable units the same controller learns the same task both faster and better, however, other reaching tasks perform less well, proportional to dissimilarity; we therefore propose that modifications enabling emergence of a more generic set of synergies are required. Finally, we propose a continuous controller for the robot, based on model predictive control, incorporating our model as a predictive component for state estimation, delay-compensation and planning, including merging of the robot and sensed environment into a single model. We test the delay compensation mechanism by controlling a second copy of the model acting as a proxy for the real robot, finding that performance is significantly improved if a precise degree of compensation is applied and show how rapidly an un-compensated controller fails as the model accuracy degrades.

The temporal structure of neural oscillations has become a widespread hypothetical \mechanism" to explain how neurodynamics give rise to neural functions. Despite the great number of empirical experiments in neuroscience and mathematical and computa- tional modelling investigating the temporal structure of the oscillations, there are still few systematic studies proposing dynamical explanations of how it operates within closed sensorimotor loops of agents performing minimally cognitive behaviours. In this thesis we explore this problem by developing and analysing theoretical models of evolutionary robotics controlled by oscillatory networks. The results obtained suggest that: i) the in- formational content in an oscillatory network about the sensorimotor dynamics is equally distributed throughout the entire range of phase relations; neither synchronous nor desyn- chronous oscillations carries a privileged status in terms of informational content in relation to an agent's sensorimotor activity; ii) although the phase relations of oscillations with a narrow frequency difference carry a relatively higher causal relevance than the rest of the phase relations to sensorimotor coordinations, overall there is no privileged functional causal contribution to either synchronous or desynchronous oscillations; and iii) oscilla- tory regimes underlying functional behaviours (e.g. phototaxis, categorical perception) are generated and sustained by the agent's sensorimotor loop dynamics, they depend not only on the dynamic structure of a sensory input but also on the coordinated coupling of the agent's motor-sensory dynamics. This thesis also contributes to the Coordination Dynam- ics framework (Kelso, 1995) by analysing the dynamics of the HKB (Haken-Kelso-Bunz) equation within a closed sensorimotor loop and by discussing the theoretical implications of such an analysis. Besides, it contributes to the ongoing philosophical debate about whether actions are either causally relevant or a constituent of cognitive functionalities by bringing this debate to the context of oscillatory neurodynamics and by illustrating the constitutive notion of actions to cognition.

Currently the central challenge facing evolutionary robotics is to determine how best to extend the range and complexity of behaviour supported by evolved neural systems. Implicit in the work described in this thesis is the idea that this might best be achieved through devising neural circuits (tractable to evolutionary exploration) that exhibit complementary functional characteristics. We concentrate on two problem domains; locomotion and sequence learning. For locomotion we compare the use of GasNets and other adaptive networks. For sequence learning we introduce a novel connectionist model inspired by the role of dopamine in the basal ganglia (commonly interpreted as a form of reinforcement learning). This connectionist approach relies upon a new neuron model inspired by notions of energy efficient signalling. Two reward adaptive circuit variants were investigated. These were applied respectively to two learning problems; where action sequences are required to take place in a strict order, and secondly, where action sequences are robust to intermediate arbitrary states. We conclude the thesis by proposing a formal model of functional integration, encompassing locomotion and sequence learning, extending ideas proposed by W. Ross Ashby. A general model of the adaptive replicator is presented, incoporating subsystems that are tuned to continuous variation and discrete or conditional events. Comparisons are made with Ross W. Ashby's model of ultrastability and his ideas on adaptive behaviour. This model is intended to support our assertion that, GasNets (and similar networks) and reward adaptive circuits of the type presented here, are intrinsically complementary. In conclusion we present some ideas on how the co-evolution of GasNet and reward adaptive circuits might lead us to significant improvements in the synthesis of agents capable of exhibiting complex adaptive behaviour.

O reconhecimento aproximado de cadeias de texto é um problema recorrente em diversas aplicações onde o computador é utilizado como meio de processamento de uma massa de dados sujeita a imprecisões, erros e distorções. Existem inúmeras metodologias, técnicas e métricas criadas e empregadas na resolução deste tipo de problema, mas a maioria delas é inflexível em pelo menos um dos seguintes pontos: arquitetura, métrica utilizada para aferir o erro encontrado ou especificidade na aplicação. Esse trabalho propõe e analisa a utilização dos Autômatos Finitos Nebulosos para a resolução desse tipo de problema. A teoria nebulosa oferece uma base teórica sólida para o tratamento de informações inexatas ou sujeita a erros, enquanto o modelo matemático dos autômatos finitos é uma ferramenta consolidada para o problema de reconhecimento de cadeias de texto. Um modelo híbrido não só oferece uma solução flexível para a resolução do problema proposto, como serve de base para a resolução de inúmeros outros problemas que dependem do tratamento de informações imprecisas. / The approximate string matching problem is recurring in many applications where computer is used to process imprecise, fuzzy or spurious data. An uncountable number of methods, techniques and metrics to solve this class of problem are available, but many of them are inflexible at least in one of following: architecture, metric or application specifics. This work proposes and analyzes the use of Fuzzy Finite State Automata to solve this class of problems. The fuzzy theory grants a solid base to handle imprecise or fuzzy information; the finite state automata is a classic tool in string matching problems. A hybrid model offers a flexible solution for this class of problem and can be a base for other problems related with imprecise data processing.

O reconhecimento aproximado de cadeias de texto é um problema recorrente em diversas aplicações onde o computador é utilizado como meio de processamento de uma massa de dados sujeita a imprecisões, erros e distorções. Existem inúmeras metodologias, técnicas e métricas criadas e empregadas na resolução deste tipo de problema, mas a maioria delas é inflexível em pelo menos um dos seguintes pontos: arquitetura, métrica utilizada para aferir o erro encontrado ou especificidade na aplicação. Esse trabalho propõe e analisa a utilização dos Autômatos Finitos Nebulosos para a resolução desse tipo de problema. A teoria nebulosa oferece uma base teórica sólida para o tratamento de informações inexatas ou sujeita a erros, enquanto o modelo matemático dos autômatos finitos é uma ferramenta consolidada para o problema de reconhecimento de cadeias de texto. Um modelo híbrido não só oferece uma solução flexível para a resolução do problema proposto, como serve de base para a resolução de inúmeros outros problemas que dependem do tratamento de informações imprecisas. / The approximate string matching problem is recurring in many applications where computer is used to process imprecise, fuzzy or spurious data. An uncountable number of methods, techniques and metrics to solve this class of problem are available, but many of them are inflexible at least in one of following: architecture, metric or application specifics. This work proposes and analyzes the use of Fuzzy Finite State Automata to solve this class of problems. The fuzzy theory grants a solid base to handle imprecise or fuzzy information; the finite state automata is a classic tool in string matching problems. A hybrid model offers a flexible solution for this class of problem and can be a base for other problems related with imprecise data processing.

Maximum linear feedback shift registers (LFSRs) based on primitive polynomials are commonly used to generate maximum length sequences (m-sequences). An m-sequence is a pseudorandom sequence that exhibits ideal randomness properties like balance, run and autocorrelation but has low linear complexity. One-dimensional Cellular Automata (1D CA) have been used to generate m-sequences and pseudorandom sequences that have high linear complexity and good randomness. This thesis considers the use of two-dimensional Cellular Automata (2D CA) to generate m-sequences and psuedorandom sequences that have high linear complexity and good randomness. The properties of these sequences are compared with those of the corresponding m-sequences and the best sequences generated by 1D CAs. / Graduate