DPhil theses

DPhil theses

The list below includes DPhil theses from the Sussex Department of Informatics (formerly Cognitive and Computing Sciences) and School of Life Sciences (formerly Biological Sciences) in the EASy area.

[B]



• Andy Balaam [Abstract] [Full Text]
  Exploring Developmental Dynamics in Evolved Neural Network Controllers.
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2006.
  

  Two new developmental neural network controllers for evolutionary robotics are described. These controllers present opportunities for experiments into many areas of development: the results of several experiments are described. Relevant former work in evolutionary robotics and the natural sciences is reviewed. The motivations of the work are presented: to generate working examples of systems which may be said to exhibit development for use as tools in its study, and to explore various open questions involving development using these tools. Several open questions are presented, most of which involve suggestions that controllers which are constrained to be developmental may offer advantages (in terms of how easily they may be evolved to exhibit certain behaviours) over more traditional controllers. The experimental work is divided into two parts. The rst part describes a controller which has neurons located in a two-dimensional space growing during the robot's lifetime in processes affected by the experience of the robot as well as its genotype. The second piece of experimental work describes a controller designed to put into practice the lessons learned in the rst part: one which has a xed size but allows for the growth and death of synapses as well as plastic changes in their weights. Little evidence was found to support the ideas behind many of the open questions explored: in particular, the controllers with added developmental dynamics were not found to have any advantage over standard controllers in tasks involving exibility to predictable changes or adaptability to different environments. Discussion is offered of the reasons for the results found, the utility of the tools developed and lessons learned about evolving developmental controllers. Possible avenues for future work are suggested.

• Pedro Paulo Balbi de Oliveira
  An empirical exploration of computations with a cellular-automata-based artificial life world.
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1994.
• Lionel Barnett [Abstract] [Full Text]
  Artificial Evolution with Neutral Networks.
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2002.
  

  In the field of search and optimisation every theorist and practitioner should be aware of the socalled No Free Lunch Theorems (Wolpert & Macready, 1997) which imply that given any optimisation algorithm, should that algorithm perform better than random search on some class of problems, then there is guaranteed to exist another class of problems for which the same algorithm performs worse than random search. Thus we can say for certain that there is no such thing as an effective “general purpose” search algorithm. The obverse is that the more we know about a class of problems, the better equipped we are to design effective optimisation algorithms for that class. This thesis addresses a quite specific class of optimisation problems - and optimisation algorithms. Our approach is to analyse statistical characteristics of the problem search space and thence to identify the algorithms (within the class considered) which exploit these characteristics - we pay for our lunch, one might say. The class of optimisation problems addressed might loosely be described as correlated fitness landscapes with large-scale neutrality; the class of search algorithms as evolutionary search processes. Why we might wish to study these problems and processes is discussed in detail in the Introduction. A brief answer is that problems of this type arise in some novel engineering tasks. What they have in common is huge search spaces and inscrutable complexity arising from a rich and complex interaction of the designed artifact with the “real world” - the messy world, that is, outside our computers. The huge search spaces and intractable structures - and hence lack of obvious design heuristics - suggests a stochastic approach; but “traditional” stochastic techniques such as Genetic Algorithms have frequently been designed with rather different search spaces in mind. This thesis examines how evolutionary search techniques might need to be be re-considered for this type of problem.

• Guillaume Barreau
  The evolutionary consequences of redundancy in natural and artificial genetic codes
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1997.
• Pete de Bourcier
  Synthetic behavioural Ecology
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1997.
• Seth Bullock [Abstract] [Full Text]
  Evolutionary Simulation Models: On their character and application to problems concerning the evolution of natural signalling systems
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1998.
  

  Evolutionary simulation modelling is presented as a methodology involving the application of modelling techniques developed within the artificial sciences to evolutionary problems. Although modelling work employing this methodology has a long and interesting history, it has remained, until recently, a relatively underdeveloped practice, lacking a unifying theoretical framework. Within this thesis, evolutionary simulation modelling will be defined as the use of simulations, constructed under constraints imposed by evolutionary theories, to explore the adequacy of these theories, through the modelling of an adaptive system’s ongoing evolution. Evolutionary simulation models may be considered to lie within the field of artificial life, since its concerns include theories of life, evolution, dynamical systems, and the relationship between artificial and natural adaptive systems. Simultaneously, evolutionary simulation modelling should be regarded as distinct from, yet complementing, existing evolutionary modelling techniques within the biological sciences. The ambit of evolutionary simulation modelling includes those systems towards which one is able to take the evolutionary perspective, i.e., systems comprising agents which change over time through the action of some adaptive process. This perspective is broad, allowing evolutionary simulation models to address linguistic models of glossogenetic change, anthropological models of cultural development, and models of economic learning, as well as models of biological evolution. Once this methodology has been defined, it is applied to a group of problems current within theoretical biology, concerning the evolution of natural signalling systems. The ubiquity of natural communication is a well attested phenomenon. However, recently the utility of such communication within a world populated by neo-Darwinian selfish individuals has been questioned. Theoretical models proposed to account for the existence of signalling within the animal kingdom are reviewed, and evolutionary simulation models are constructed in an attempt to assess these theories. Specifically, models of the evolution of complex symmetry, and models of the evolution of honesty, are addressed.

[C]



• Dave Cliff
  Animate Vision in an Artificial Fly: A study in Computational Neuroethology
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1991.
• Paulo Costa
  Explorations in Insect Sociality: Towards a Unified Approach
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1997.

[D]



• Kim Dae Gyu
  Mapping based constraint handling methods for evolutionary algorithms
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2000.
• Kyran Dale
  When Worlds Collide: some simple models of navigation using vision and a compass
  DPhil Thesis. School of Biological Sciences, University of Sussex. 2001.
• Robert Davidge
  Computer Processors which behave like Unicellular Organisms: A thesis in Artificial Life
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1994.
• Ezequiel Di Paolo [Abstract] [Full Text]
  On the Evolutionary and behavioural Dynamics of Social Coordination: Models and Theoretical Aspects
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1999.
  

  An exploration is presented of the interplay between the situated activity of embodied autonomous organisms and the social dynamics they constitute in interaction, with special emphasis on evolutionary, ecological and behavioural aspects. The thesis offers a series of theoretical and methodological criticisms of recent investigations on the biology of social behaviour and animal communication. An alternative theoretical framework, based on a systemic theory of biological autonomy, is provided to meet these criticisms and the elaboration of the corresponding theoretical arguments is supported by the construction and analysis of mathematical and computational models.
  A game of action coordination is studied by a series of game-theoretic, ecological and computational models which, by means of systematic comparisons, permit the identification of the evolutionary relevance of different factors like finite populations, ecological and genetic constraints, spatial patterns, discreteness and stochasticity. Only in an individual-based model is it found that cooperative action coordination is evolutionarily stable. This is due to the emergence of spatial clusters in the spatial distribution of players which break many of the in-built symmetries of the game and act as invariants of the dynamics constraining the path of viable evolution.
  An extension to this model explores other structuring effects by adding the possibility of parental influences on phenotypic development. The result is a further stabilization of cooperative coordination which is explained by the presence of self-promoting networks of developmental relationships which enslave the evolutionary dynamics.
  The behavioural aspects involved in the attainment of a coordinated state between autonomous systems are studied in a simulated model of embodied agents coupled through an acoustic medium. Agents must locate and approach each other only by means of continuous acoustic signals. The results show the emergence of synchronized rhythmic signalling patterns that resemble turn-taking which is accompanied by coherent patterns of movement. It is demonstrated that coordination results from the achievement of structural congruence between the agents during interaction.
  

• Stephen Dunn
  Modelling the Neural Network Underlying Feeding behaviour in the Snail Lymnaea Stagnalis
  DPhil Thesis, 1998.

[F]



• Joe Faith [Abstract] [Full Text]
  Emergent Representations: Dialectical Materialism and the Philosophy of Mind.
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2000.
  

  If minds are the products of brains (and they are) then surely neuroscience should tell us something about the philosophy of mind? And, if minds are also the products of natural and social history (and they are) then surely the study of these subjects will tell us something about mind too? I am not the first to suggest these lines of inquiry but they have both traditionally lead to the same conclusion, namely a virulent scepticism about our ability to know the world. We seem to be stuck in a Faustian bargain in which we gain scientific knowledge at the expense of philosophical doubt.
  This thesis is an attempt to break this bargain, in which I start from the conviction that we can know the world, and then ask what kind of science, both natural and social, can make sense of this ability. We do not just need a philosophy of mind that fits our science, we also need a science that fits our philosophy of mind. We must fiddle with both sides of the equation in order to get a fit. In the course of this fiddling I challenge reductionist and empiricist assumptions about science, I question the philosophical tradition that dates back to Frege's `linguistic turn', and I draw parallels between Marx's theory of history and Darwin's theory of natural selection. The result is a realist philosophy of mind that is built on our ability to interact with and change the world, rather than on our ability to contemplate it passively.

• Chrisantha Fernando [Abstract] [Full Text]
  The Evolution of the Chemoton.
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2006.
  

  Tibor Ganti proposed the chemoton as the minimal system capable of open-ended evolution, consisting of three stoichiometrically coupled autocatalytic systems, a metabolism, a membrane and a template replicating system. Our models show that an autocatalytic metabolic system could evolve only if chemical and physical niches could be discovered that limited the extent of tapping side-reactions. After that, selection at the protocell level would have favored those containing clean template replication rather than messy polymerization. Template length could additionally confer a weak Lamarkian inheritance mechanism. We propose a mechanism for the origin of long template replication. Although unlimited nucleic acid elongation is possible at low temperatures, elongation out-competes replication for limited monomer resource. However, in a reactor with a low temperature baseline, with denaturation spikes, the minimal functionality required of an RNA motif capable of self-replication is that it can `cut itself out' of an elongating strand, and accept oligomers upon renaturation. No ligase function or processivity is necessary, or possible. Several artificial ribozymes were designed and tested. Such a restriction ribozyme function is readily observed in extant RNAs, may have been selected for, and is an ideal mechanism for dealing with elongation side-reactions.

• Paul Graham
  Visually guided route navigation in the wood ant (Formica rufa)
  DPhil Thesis. School of Biological Sciences, University of Sussex. 2002.

[H]



• Inman Harvey [Abstract] [Full Text]
  The Artificial Evolution of Adaptive behaviour
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1995.
  

  A methodology is presented for the design through artificial evolution of adaptive complex systems, such as the control systems of autonomous robots.
  Genetic algorithms have largely been tailored towards optimisation problems with a fixed and well-defined search-space; the SAGA (Species Adaptation Genetic Algorithms) framework is introduced for the different domain of long term artificial evolution, where the task domain is ill-defined and can increase in complexity indefinitely. Genotypes should be able to increase in length indefinitely, and evolution will take place in a genetically converged population. Significant changes from normal genetic algorithm practice follow from this.
  It is shown that changes in genotype length should be restricted to gradual ones. Appropriate mutation rates are proposed to encourage exploration of the high-dimensional fitness landscape without losing gains already made. Tournament selection, or similar ranking methods, are advocated as a way of maintaining selection pressures at a known rate. A crossover algorithm is introduced, which allows for recombination of genotypes of different lengths without undue confusion. The significance of a developmental process from genotype to phenotype, of co-evolution and of neutral drift through genotype space, are discussed.
  As a class of control systems appropriate for evolution, programming languages are dismissed in favour of realtime dynamical recurrent connectionist networks; issues of time, representation and learning in such networks are discussed. A whole complex system, comprised of such a network together with sensory and motor systems, is characterised as a dynamical system with internal state, coupled to a dynamical environment.
  Applications of these theoretical frameworks of artificial evolution and of control systems are demonstrated in a series of experiments with mobile robots engaged in navigational tasks using low-bandwidth sensors. Initial experiments are in simulation; the validity of such simulations and the significance of noise is discussed. Then experiments move to a real-world domain, with the use of a specialised piece of hardware which allows the automatic evaluation of populations of mobile robots using real low-bandwidth vision to navigate in a test environment. Evolution of capabilities is demonstrated in a sequence of navigational tasks of increasing complexity.

• Horst Hendriks-Jansen
  Situated Activity, Interactive Emergence, and Human Thought
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1994.
• Rainer Hilscher [Abstract] [Full Text]
  Agent based models of competitive speciation
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2005.
  

  Competitive speciation is a complex evolutionary phenomenon that is result of many interacting factors. Feeding strategies, mate search tactics of females, and ecology influence the dynamics of competitive speciation. This dissertation explores novel aspects of competitive speciation that have not been addressed in existing models. Speciation due to competitive interactions alone is shown to be inherently instable. If conditions for competition change so does the resultant species system.
  The traditional 2-phase pattern view of sympatric speciation of a stable polymorphism followed by the evolution of assortative mating is shown to only apply to an environment with 2 abutting habitats. In an environment with a continuous resource distribution divergence of the population and evolution of assortative mating occur simultaneously. In existing models assortative mating and reproductive isolation are implicitly considered equivalent. In this dissertation it is found that the evolution of assortative mating does not automatically imply reproductive isolation. Sequential speciation events frequently occur and the number of species in a given environment is strongly influenced by the feeding strategy of individuals. Assortative mating plays a pivotal role in any sympatric speciation event but has so far been neglected as a research topic.
  Assortative mating is being studied here by comparing three mate search tactics. A hybrid tactic where females switch from threshold to best-of- n mating if they fail to find a mate with the default threshold tactic proves to be the most facilitative for speciation. Finally, competition in n -dimensional resource space has so far received very little attention. With the simulation platform developed for the present dissertation n -dimensional competition can be investigated. N -space competitive speciation is shown to be very constrained. Higher dimensional traits do not imply more species and independent or modularized dimensions inhibit population divergence.
  This dissertation introduces a modular Agent Based Modeling simulation platform. Such a simulation methodology makes it possible to address a wide range of research questions without compromising the inner logic of implemented models. The present modeling platform allows for a direct comparison of results since all simulations are based on the same model assumptions.

[J]



• Nick Jakobi [Abstract] [Full Text]
  Minimal Simulations For Evolutionary Robotics
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1998.
  

  For several years now, various researchers have endeavoured to apply artificial evolution to the automatic design of control systems for robots. One of the major challenges they face is how the fitness of evolving controllers should be tested when each evolutionary run typically involves hundreds of thousands of such assessments. This thesis puts forward new techniques for evolving control systems for real robots using easy-to-build, fast-running simulations. It begins with a tutorial in state-of-the-art Evolutionary Robotics and discusses the best types of neural network, encoding scheme, genetic algorithm and genetic operators to use - and how to use them. Several novel types are introduced, and their relative merits over previous approaches are discussed. After analysing the conditions that must be met by controllers if they are to perform the same behaviour in reality as they do in simulation, a new methodology is proposed for building minimal simulations within which controllers that meet these conditions will successfully transfer into reality. Techniques are then put forward for forcing controllers that evolve to be reliably fit within such minimal simulations to meet these conditions. Four sets of experiments are reported, all involving minimal simulations. Controllers were evolved for a small mobile robot that could solve a T-maze in response to a light cue, target recognition and approach behaviours were evolved for a visually guided mobile robot, walking and obstacle-avoiding behaviours were evolved for an eight-legged robot and motion-tracking behaviours were evolved for a simple panning camera head. In all four cases, the evolution of complex robot behaviours that would have taken many months to evolve if fitness evaluations had been performed in reality was performed in a matter of hours, and controllers that evolved to be reliably fit in simulation displayed extremely robust behaviour when downloaded into reality.

• Tudor Jenkins [Abstract] [Full Text]
  Evolutionary Robotics' First Words
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2006.
  

  For just over a decade great strides have been made with the situated bottom up approach to artificial intelligence yet there are many problem domains still much more suited to the traditional top down artificial intelligence techniques that it seeks to replace. This thesis considers how the acquisition of a grounded symbol system suited for communication within one of these bottom up approaches, evolutionary robotics, offers a way of integrating some of the beneficial techniques of traditional AI with the new bottom up approach. Evolving linguistically capable robots is a very multifaceted problem and a categorization of the various domains is made through a detailed exploration of existing work within related fields. Whilst the acquisition of symbolic signaling systems within this bottom up domain has received some attention, it has been restricted to trivial search spaces or the use of high level components. These fail to allow for an atomic approach to the development of signals integrated at a base level with sensors and behaviours to allow for an infinite range of possible signaling strategies necessary if we are to achieve open language. Integration of transmission and reception strategies presents problems of finding solutions within huge search spaces within this approach and three methods to limit this space are proposed. Each of these offers some form of compromise between the rate at which the signaling can arise and the range of problem domains to which it can be applied. Experiments are undertaken to assess the three different methods, which rely on imitation, babbling and forced externalization of self-control.

• Ibrahim Kuscu
  Genetic based methods for hard learning problems
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1999.
  

[L]



• Paul Layzell [Abstract] [Full Text]
  Hardware Evolution: On the Nature of Electronic Circuits Derived Through Artificial Evolution
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2001.
  

  Artificial evolution is capable of deriving electronic circuits of very different nature to those designed using established top-down methodologies. Such circuits may have no clear functional decomposition, and can rely heavily on subtle ‘parasitic’ properties not normally exploited in conventional design. They can exhibit more efficient component usage, but less tolerance to fabrication and environmental variations. If an evolved circuit’s operation is understood, steps can be taken to increase its tolerance to these factors, thereby enhancing its utility. However, analysis is problematic if the circuit possesses complex dynamics dependent on unknown parasitic properties. The topology and operation of evolved circuits are derived incrementally from the evolutionary process. Analysis is hence facilitated if established reverse-engineering techniques are also applied to ancestor circuits normally considered evolutionary by-products. But this is difficult and laborious for circuits evolved directly on commercial reconfigurable FPGA chips, which have little or no accessibility to internal nodes for probing. As well as impeding analysis, FPGAs offer little choice of circuit primitives, restricted interconnection architecture, and most are susceptible to self-destruction. A new, less restrictive configurable research platform is presented, which possesses a comprehensive interconnection architecture and circuit primitives implemented as plug-in daughterboards, allowing a huge variety to be hosted with direct access to their pins for probing. The platform cannot be destroyed by illegal configurations if certain basic rules are followed. A series of experiments demonstrates the platform’s capacity to aid research of fundamental issues dominating hardware evolution, yielding useful insights on analysis, genotype encoding, choice of basic elements, portability, evolved topologies, exploitation of configuration circuitry, and extrinsic versus intrinsic evolution. The platform is then used to show that populations of evolved circuits contain individuals inherently tolerant to major faults. Through applying automatic analysis of ancestor circuits, the underlying mechanism responsible is revealed.

• Ronald Lemmen
  Towards a Non-Cartesian Cognitive Science in the light of the philosophy of Merleau-Ponty
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1998.
• Giles Mayley
  Explorations into the interactions between learning and evolution using algorithms
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2000.
• Jason Noble [Abstract] [Full Text]
  The Evolution of Animal Communication Systems: Questions of Function Examined through Simulation
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1998.
  

  Simulated evolution is used as a tool for investigating the selective pressures that have influenced the design of animal signalling systems. The biological literature on communication is first reviewed: central concepts such as the handicap principle and the view of signalling as manipulation are discussed. The equation of "biological function" with "adaptive value" is then defended, along with a workable definition of communication. Evolutionary simulation models are advocated as a way of testing the coherence of a given theory. Contra some ALife enthusiasts, simulations are not alternate worlds worthy of independent study; in fact they fit naturally into a Quinean picture of scientific knowledge as a web of modifiable propositions. Existing simulation work on the evolution of communication is reviewed: much of it consists of simple proofs of concept that fail to make connections with existing theory. A particular model (MacLennan and Burghardt, 1994) of the evolution of referential communication in a co-operative context is replicated and critiqued in detail.
  Evolutionary simulations are then presented that cover a range of ecological scenarios; the first is a general model of food- and alarm-calling. In such situations signallers and receivers can have common or conflicting interests; the model allows us to test the idea that a conflict of interests will lead to an arms race of ever more costly signals, whereas common interests will result in signals that are as cheap as possible. The second model is concerned with communication during aggressive interactions. Many animals use signals to settle contests, thus avoiding the costs associated with fighting. Conventional game-theoretic results suggest that the signalling of aggression or of strength will not be evolutionarily stable unless it is physically unfakeable, but some recent models imply that cost-free, arbitrary signals can be reliable indicators of both intent and ability. The simulation, which features continuous-time perception of the opponent's strategy, is an attempt to settle the question. The third model deals with sexual signalling, i.e., elaborate displays that are designed to persuade members of the opposite sex to mate. The results clarify the question of whether such displays are the pointless result of runaway sexual selection, or whether they function as honest and costly indicators of genetic quality.
  The models predict the evolution of reliable communication in a surprisingly narrow range of circumstances; a serious gap remains between these predictions and the ethological data. Future directions for simulation work are discussed.
  

[O]



• Gabriella Ochoa [Abstract] [Full Text]
  Error thresholds and optimal mutation rates in genetic algorithms
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2001.
  

  When applying a genetic algorithm to solve a given problem, the designer faces a large number of choices, with little theoretical guidance and few rules of thumb about how to proceed. Among these choices, the setting of evolutionary parameters (e.g. mutation rate, recombination rate, pop­ ulation size and selection parameters) is important since their values determine the performance of the algorithm to a great extent. However, finding a good combination of parameters is not an easy task since they interact with one another non­linearly and cannot be optimised one at a time. Moreover, `optimal' parameter settings are believed to be problem­dependent. The mutation rate is acknowledged as one of the most sensitive parameters, so good heuristics for setting the mutation rate are welcomed. This thesis brings the fundamental notion of the error thresholds of replication from molec­ ular evolution into the field of evolutionary computation. Error thresholds are intuitively related to the idea of an optimal balance between exploration and exploitation in genetic search. So, it is hypothesised and empirically demonstrated here, that error thresholds are related to the more familiar notion of optimal mutation rates in GAs. This finding sheds new light on the sensitivity of the mutation rate and points toward useful heuristics for setting this parameter. Some results on the effects and usefulness of recombination are also presented. This dissertation also introduces consensus sequence plots, which are adapted from theoretical biology, as a new visualisation tool to the genetic algorithms community. They are used for locating error thresholds on general land­ scapes, and are shown to reveal several features of the landscape structure. The insights and empirical evidence gathered here support a heuristic that sets a rate based on one mutation per genotype, to be scaled according to the selection pressure and also potentially modified for very redundant genotypes. However, since the selection pressure can be controlled, this rule is shown to hold over a wide range of problem types.

• Ka Yin Caroline Ong [Abstract] [Full Text]
  Philosophical Robotics
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2006.
  

  The general belief amongst researchers seeking to create/build robots that are alive/lifelike within Robotics and A-Life is that our models are either missing something fundamental about living things – ‘the juice’ [Brooks, 2003] – or, component-wise, have yet to reach the critical mass at which life emerges from the complex dynamics. This thesis addresses the project of realising certain properties associated with living organisms in autonomous robots – autonomy, intentionality, feelings and cognition – through the exploration and clarification of the relations between them as based on autopoietic theory [Maturana & Varela, 1980], and further complemented by other theories. Insights gained are then applied to the analysis of autonomous robots, with the goal of answering – to what extent, through intelligent design, can autonomous robots realise certain properties observed in biological life without being fully alive in the autopoietic sense? By focussing on the hierarchy inherent in living systems with nervous systems (neural/metabolic decoupling), as well as metabolic processes, it is possible to conceptually separate autonomy into three forms – cognitive, energetic and material. This is essentially, a semi-autonomous subsystems approach. With these three forms as foundation, it was concluded that for autonomous robots, cognitive autonomy, characterised by self-law (behavioural), is possible, as is energetic autonomy, which requires self-sufficiency in energy management. These two can also be hierarchically linked up. However, material autonomy is currently untenable. For intentionality and feelings, both being conceptually triadic in relation, the minimum required is energetic autonomy as the processes that generate the subject pole is necessary. Cognition, strictly defined as processes that are complexly neural in nature rather than purely adaptive, would require the decoupling of the neurocognitive component from the metabolic. A conceptual framework for designing/building materially embodied autonomous robots that are not fully alive in the autopoietic sense is provided.

[P]



• Alex Penn [Abstract] [Full Text]
  Ecosystem Selection: Simulation, Experiment and Theory
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2006.
  

  This thesis presents a body of simulation and experimental work on the topic of artificial ecosystem selection: selection on the level of ‘ecosystems’ or multi-species communities in order to alter ’ecosystem-level properties’. First shown to be effective in artificial selection experiments on microbial ecosystems (88; 87), the phenomenon of ecosystem selection has been demonstrated, but not the mechanisms by which it operates. The specific requirements that must be met by a population of units in order for a response to selection to occur form the guiding framework for this thesis. The concepts of variation, heritability, and phenotype are discussed in the ecosystem context, along with possible mechanisms via which ecological systems could satisfy these criteria. Particular emphasis is placed on possible novel sources of higher-level heritable variation, qualitatively different to those which exist at the individual level, some of which are demonstrated in model systems. Artificial ecosystem selection is modelled by combining Lotka-Volterra competition dynamics with a simulated ecosystem-level selection procedure. The responses to selection for both linear and non-linear ecosystem-level fitness functions are highly significant. Higher-level sources of heritable variation, relatively stable combinations of species composition and interactions, are demonstrated. In addition, it is shown that model ecosystems can respond to selection on an ecosystem-level trait without genetic change in the constituent species. Instead a limited heredity is possible via selection of different potential attractors in community composition. The dependence of the response to selection on the form of the ecological dynamics, species interactions, sample size, and form of the fitness function is discussed. The results of practical experiments on using ecosystem selection to improve the growth of L. culinaris in a semi-arid environment with degraded soil are also presented. Building upon work on producing mycorrhizal inoculum for re-vegetation projects, the experiment was carried out in basic conditions in order to assess the viability of this method as a tool for use in developing countries. Selected inoculum performed significantly better than various control treatments in the majority of lines. However no significant differences between treatments could be detected when tested in a field trial. Issues arising from work in artificial ecosystem selection are extensively discussed. Relevant literature is reviewed, connecting ecosystem selection to work in major transitions in evolution, group selection, and the origins and meaning of individuality in biological systems. Finally, consideration is extended to the possibility of ecosystem selection occurring in natural settings. Cases in which ecosystem selection might be occurring are reviewed. New and further directions for research in the area of ecosystem selection are discussed.

• Andy Philippides [Abstract] [Full Text]
  Modelling the Diffusion of Nitric Oxide in Brains
  DPhil Thesis. School of Biological Sciences, University of Sussex. 2001.
  

  This thesis investigates modelling the diffusion of the gaseous transmitter nitric oxide (NO) in brains in order to study its role in neuromodulation and aims to address issues in both computational neuroscience and evolutionary robotics. It begins with a review of previous modelling and shows that the techniques used, while a reasonable first approximation, are limited in a number of ways. New methods are then introduced which, for the first time, allow accurate morphological modelling of arbitrarily shaped NO sources in the brain with two main effects, the ‘reservoir’ and ‘delay’ effects, highlighted. This analysis is then extended to incorporate diffusion of NO from multiple small sources and shows how such sources, while unable to signal while acting alone due to their small size, can together produce an effective ‘volume signal’ throughout large regions of the brain. Moreover, the benefits of a volume signal that is generated by many small sources as opposed to fewer larger sources are highlighted and explained, and the implications for neural processes such as linking brain activity to increased cerebral blood flow are discussed. The next section describes an existing abstract, artificial neural network (ANN) based model of gaseous signalling in brains (the ‘Gasnet’ model) which has been successfully applied to the field of autonomous robotics. Extensions of this model based on results from the previous chapters are then detailed with the new model being applied to a real-world shape discrimination task. This problem is that which was used in conjunction with the original GasNet and which demonstrated the increased evolvability of such networks compared to traditional ANNs. An extension to the GasNet model is shown to further increase network evolvability. The thesis concludes with a discussion and some ideas for future research

[S]



• Anil Seth [Abstract] [Full Text]
  On the relations between behaviour, mechanism and environment: explorations in artificial evolution
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2000.
  

  This thesis presents an externalist exploration of the relations between behaviour, mechanism, and environment, as they arise in a variety of agent-environment systems. It offers contributions at conceptual, methodological, and empirical levels of discourse.
  Externalism describes the attempt to understand the internal in terms of the external, and the thesis begins by developing a conceptual framework justifying the use of artificial evolution models in the application of this perspective to agent-environment systems. In particular, it is argued that such models play a crucial role in elaborating the distinction between behavioural and mechanistic levels of description. There follows a series of models, of both game-theoretic and evolutionary-robotic character, which focus on explaining internal complexity in terms of adaptation to (external) environmental variability. As part of this project, accounts of the evolution of complexity in general are critiqued, and the practical importance of noise in artificial evolution is discussed.
  The thesis continues with an integration of this externalist project with the well established theoretical biology methodology of optimal foraging theory. A novel methodology - individual-based optimal situated modelling - is described, which extends orthodox optimal foraging theory through (1) the use of artificial evolution as an optimisation procedure and, (2) modelling agent-environment interaction at the level of situated perception and action. The conceptual leverage afforded by this extension is illustrated in its application to the problem of behaviour coordination in a simple agent-environment system; for example, the need for a dedicated action selection mechanism is questioned. The methodology is then addressed to a range of issues in contemporary theoretical biology and psychology: the interference function, the ideal free distribution, and the individual matching law, issues which are united by a concern with individual choice and its collective consequences. A series of models are presented which demonstrate, in these contexts, that (1) behaviours for which there is debate about the level of complexity required for their underlying mechanism, can be subserved by surprisingly simple mechanisms, and (2) behaviours which may be irrational when expressed by an isolated individual can be understood as rational in a group context.

• Oliver Sharpe [Full Text]
  Towards a Rational Methodology for using Evolutionary Search Algorithms
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2002.
  
• Nick Sharples
  Evolutionary approaches to adaptive protocol design
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2002.
• Hanson Schmidt-Cornelius [Abstract] [Full Text]
  Reverse engineering an active eye
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2002.
  

  The research presented in this dissertation investigates new concepts of biologically inspired active vision systems, using high power, linear actuators. First, fundamental vision paradigms are discussed, contrasting passive and purposive active vision. This leads to physiological and historical investigations into extra-oculomotor mechanisms and laws that define human eyeball positioning. Based on this background, the design of a monocular extra-oculomotor robot is devised that combines a parallel architecture, static balancing and opposing actuation. This design is then implemented, resulting in the construction of an experimental active eye environment. The mechanical, electrical and software development phases are discussed. The unique control and mechanical properties of the active vision system here necessitate a range of low level control algorithms that are normally not required for conventional vision systems with stepper, servo or direct drive motor control. Three biologically based control models are used to form the basis of a range of adaptive control models that are tested on the experimental environment of the mechanical eye. The controllers developed here perform similar operations to those expected from the control layers of biological vision systems. This is especially the case for the neural control model of the superior colliculus, which has not been tested in the context of such an environment before. Experimental results show that this new approach to artificial active vision systems is viable, replacing position controlled actuation with dynamic mechanisms, regulated by adaptive controllers.

• Carol Shergold
  Sensory-motor coordination: adapting to disruptions
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2001.
• Tom Smith
  The Evolvability of Artificial Neural Networks for Robot Control
  DPhil Thesis. School of Biological Sciences, University of Sussex. 2002.
• Terry Stewart [Abstract]
  Learning in Artificial Life: conditioning, concept formation, and sensorimotor loops
  MPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2002.
  

  The first half of this thesis is a review of neural models of associative learning, with a particular focus on two things: the ability to form 'concepts' (extracting patterns from the sensory data) and the capability of dealing with embodied agents.... The second half details a closer examination of distributed adaptive control, in light of the previous discussion. A series of experiments are performed on a re-implementation of this learning algorithm which compare its associative learning characteristics to those of the most basic of natural associative learning methods: classical conditioning.... The result is that while distributed adaptive control may show the surface capabilities of classical conditioning (the ability to have a conditioned stimulus act as a predictor for an unconditioned stimulus), it does not have the deeper abilities of classical conditioning....

[T]



• Adrian Thompson
  Hardware Evolution: Automatic design of electronic circuits in reconfigurable hardware by artificial evolution.
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1996.
• Elio Tuci [Abstract] [Full Text]
  An exploration on the evolution of learning behaviour using robot based models
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2004.
  

  The work described in this thesis concerns the study of the evolution of simple forms of learning behaviour in artificial agents. Our interest in the phylogeny of learning has been developed within the theoretical framework provided by the “ecological approach” to the study of learning. The latter is a recent theoretical and methodological perspective which, contrary to that suggested by the classical approaches in animal and comparative psychology, has reconsidered the importance of the evolutionary analysis of learning as a species- niche-specific adaptive process, which should be investigated by employing the conceptual apparatus originally developed by J.J. Gibson within the context of visual perception. However, it has been acknowledged in the literature that methodological difficulties are hindering the evolutionary ecological study of learning. We argue that methodological tools —i.e., artificial agent based models —recently developed within the context of biologically-oriented cognitive science can potentially represent a complementary methodology to investigate issues concerning the evolutionary history of learning without losing sight of the complexity of the ecological perspective. Thus, the experimental work presented in this thesis contributes to the discussion on the adaptive significance of learning, through the analysis of the evolution of simple forms of associative learning in artificial agents. Part of the work of the thesis focuses on the study of the nature of the selection pressures which facilitate the evolution of associative learning. The results of these simulations suggest that ecological factors might prevent the selection from operating in favour of those elements of the “learning machinery” which, given the varying nature of the environment, are of potential benefit for the agents. Other simulations highlight the properties of the agent control structure and the characteristics of particular features of the ecology of the learning scenario which facilitate the evolution of learning agents.

[W]



• Michael Wheeler
  The Philosophy of Situated Activity
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1995.
• Paul Wilken
  Spiking models of local neocortical circuits
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 2001.
• Rachel Wood [Abstract] [Full Text]
  Small but perfectly formed: evolutionary robotics models of 'minimal' developmental systems
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. Submitted 2006.
  

  This thesis describes work using evolutionary robotics methods to evolve `minimal' developmental systems. Evolutionary robotics has a proven track record as a methodology for automating the design of systems capable of non-trivial adaptive behaviour. In addition evolutionary robotics has been used by a number of researchers as a means to investigate the evolution of adaptive behaviour in natural systems. In this thesis evolutionary robotics methods are adopted for the exploration of the development of adaptive behaviour.
  This approach is founded on the notion that developing agents and their contexts comprise systems characterised by complex dynamic interactions instantiated at multiple timescales. A discussion is presented of the evidence from psychology and biology which has informed the perspective advanced here; specifically that developing agents are best understood as systems and that the context in which much of the development observed in higher animals occurs is fundamentally social.
  The thesis also includes discussion of the adaptive systems and evolutionary robotics research which has contributed to the methodology proposed in this work. Three pieces of empirical work using this methodology are presented. In the first two of these experiments the use of evolutionary robotics as a method for obtaining `minimal' developmental systems is assessed. Results of these investigations indicate that manipulation of context can be used to obtain developmental systems which extend beyond the boundary of the developing individual.
  The third set of experiments described here takes a slightly different approach, asking if this methodology can be used to evolve a specific developmental trajectory, i.e. the A not B error observed in delayed manual search tasks with human infants. This model demonstrates successful performance of the delayed manual search task and includes investigation of the role of homeostatic dynamics in perseverative task errors.
  In this work the value of the proposed approach for modelling complex empirical data is assessed. The thesis concludes with proposals for further development of this approach and discussion of the potential for `generative' modelling in which minimal developmental system models are used to suggest new experimental manipulations in natural adaptive systems.

• Andy Wuensche
  Attractor Basins of Discrete Networks: Implications on self-organisation and memory
  DPhil Thesis. School of Cognitive and Computing Sciences, University of Sussex. 1997.
• Ricardo Salem Zebulum
  Evolutionary Electronics
  Doctor in Electrical Engineering Sciences at the Catholic University of Rio, Brazil (with much of the research conducted within the EASy group), 1999.