Advances in Artificial Life

The paper considers supervised learning algorithm of nonlinear perceptron with dynamic targets adjustment which assists in faster learning and cognition. A difference between targets of the perceptron corresponding to objects of the first and second categories is associated with stimulation strength. A feedback chain that controls the difference between targets is interpreted as synthetic emotions. In a population of artificial agents that ought to learn similar pattern classification tasks, presence of the emotions helps a larger fraction of the agents to survive. We found that optimal level of synthetic emotions depends on difficulty of the pattern recognition task and requirements to learning quality and confirm Yerkes-Dodson law found in psychology.

Jakob von Uexküll’s theory of the

Umwelt

is described and it is used to show how perceptual states can be defined. It is described how perceptual cues are selected over evolutionary time and defined by the organism that experiences them. It is then argued that by applying the model of the

Umwelt

to describe an animat’s behaviour, we can model the normally distributed, dynamic activations of the animat as discrete perceptual states.

We present two original computational models — globular universe and autopoietic automata — capturing the basic aspects of an evolution: a construction of self–reproducing automata by self–assembly and a transfer of algorithmically modified genetic information over generations. Within this framework we show implementation of autopoietic automata in a globular universe. Further, we characterize the computational power of lineages of autopoietic automata via interactive Turing machines and show an unbounded complexity growth of a computational power of automata during the evolution. Finally, we define the problem of sustainable evolution and show its undecidability.

The contribution provides an example of how a formal model of some life-like functions – the so called eco-grammar (EG) system – provides a framework in which it is formally provable that the computational power of the model – under some very natural circumstances derived from the specificities of living systems, esp. from their embodiment – may overcome the computational limits of traditional computing models, perhaps also the computational power of the universal Turing machine.

In this paper, we propose a formal definition of the perception as a behavioral dynamical attraction basin. The perception is built from the integration of the sensori-motor flow. Psychological considerations and robotic experiments on an embodied “intelligent” system are provided to show how this definition can satisfy both psychologist and robotician point of view.

The

strong continuity thesis

was introduced into the artificial life literature in 1994, [5], but since then has not received the attention and further development it merits. In this paper, I explain why if we are to identify genuine continuity between life and mind, a shift in perspective is needed from thinking about living and minded things and processes, to thinking about Life itself and Mind itself. I describe both life and mind as self-preserving processes and argue that this notion accounts for their purported continuity, drawing on research in embedded and embodied cognition to make my case. I then respond to Peter Godfrey-Smith’s observation that any view on which thought requires language is inconsistent with the strong continuity thesis by arguing that although such a view of thought might be rendered consistent with the thesis, a dynamic systems approach to cognition, i.e., one wherein thought is language-

in

dependent, is much more conducive to identifying genuine life-mind continuity.

Genetic regulatory networks (GRNs) control gene expression and are responsible for establishing the regular cellular patterns that constitute an organism. This paper introduces a model of biological development that generates cellular patterns via chemical interactions. GRNs for protein expression are generated and evaluated for their effectiveness in constructing 2D patterns of cells such as borders, patches, and mosaics. Three types of searches were performed: (a) a Monte Carlo search of the GRN space using a utility function based on spatial interestingness; (b) a hill climbing search to identify GRNs that solve specific pattern problems; (c) a search for combinatorial codes that solve difficult target patterns by running multiple disjoint GRNs in parallel. We show that simple biologically realistic GRNs can construct many complex cellular patterns. Our model provides an avenue to explore the evolution of complex GRNs that drive development.

A developmental Artificial Neural Tissue (ANT) architecture inspired by the mammalian visual cortex is presented. It is shown that with the effective use of gene regulation that large phenotypes in the form of Artificial Neural Tissues do not necessarily pose an impediment to evolution. ANT includes a Gene Regulatory Network that controls cell growth/death and activation/inhibition of the tissue based on a coarse-coding framework. This scalable architecture can facilitate emergent (self-organized) task decomposition and require limited task specific information compared with fixed topologies. Only a global fitness function (without biasing a particular task decomposition strategy) is specified and self-organized task decomposition is achieved through a process of gene regulation, competitive coevolution, cooperation and specialization.

Plant morphogenesis is the development of plant form and structure by coordinated cell division and growth. We present a dynamic computational model of plant morphogenesis at cellular level. The model is based on a self-reproducing cell, which has dynamic state parameters and spatial boundary geometry. Cell-cell signalling is simulated by diffusion of morphogens, and genetic regulation by a program or script. Each cell runs an identical script, equivalent to the genome. The model provides a platform to explore coupled interactions between genetic regulation, spatio-mechanical factors, and signal transduction in multicellular organisation. We demonstrate the capacity of the model to capture the key aspects of plant morphogenesis.

Developmental models simulate the spatio-temporal development of a complex system. The system described in this paper combines the advantages of a number of previously disparate models, such as timed L-systems and cellular programming, into a single system with extensive modeling flexibility. The new system includes the ability to specify dynamic hierarchies as part of the specification, and a decoupling of cell development from interpretation. Examples in application areas of computer animation and music synthesis are provided.

Maternal influence on offspring goes beyond strict nuclear (DNA) inheritance: inherited maternal mRNA, mitochondria, caring and nurturing are all additional sources that affect offspring development, and they can be also shaped by evolution. These additional factors are called maternal effects, and their important role in evolution is well established experimentally. This paper presents two models for maternal effects, based on a genetic algorithm and simulated development of neural networks. We extended a model by Eggenberger by adding two mechanisms for maternal effects: the first mechanism attempts to replicate maternal cytoplasmic control, while the second mechanism replicates interactions between the fetus and the uterine environment. For examining the role of maternal effects in artificial evolution, we evolved networks for the odd-3-parity problem, using increasing rates of maternal influence. Experiments have shown that maternal effects increase adaptiveness in the latter model.

We introduce METAMorph, an open source software platform for the experimental design of simulated cellular development processes using genomes encoded as genetic regulatory networks (GRNs). METAMorph allows researchers to design GRNs by hand and to visualise the resulting morphological growth process. As such, it is a tool to aid researchers in developing an understanding of the expressive properties of GRNs. We describe the software and present our preliminary observations in the form of techniques for realising some common structures.

This paper focuses on the environmental role in morphogenesis in dynamic morphologies (DM). We discuss the benefits of morphological plasticity (MP) and introduce our Environment-Phenotype Map (E-P Map) framework in order to investigate and classify the

continual

development in DMs and morphologically adaptive behaviour. We present our MP-capable system the Artificial Cytoskeleton (ArtCyto), housed within our DM the ‘Cellanimat’, with an E-P Map closely based on MP examples from cell physiology. We provide experimental results to demonstrate that with this single E-P Map a bifurcation in morphology can occur, caused only by a difference in the environment, mirroring evidence from physiological data of fibroblast cell chemotaxis and macrophage cell phagocytosis.

Inspired by the recent advances in evolutionary biology, we have developed a self-organising, self-adaptable cellular system for multitask learning. The main aim of our project is to design and prototype a framework that facilitates building complex software systems in an automated and autonomous fashion. The current implementation consists of specialised programs that call (co-operate with) their local neighbours. The relationships between programs self-assemble in a symbiotic-like fashion.

Specialisation is achieved by stochastic exploration of alternative configurations and program space. A collection of global and local behaviours have been observed and investigated. Based on preliminary experimental results, certain behaviours that spontaneously exhibit self-organisation and self-assembly are discussed.

This paper is part of a larger project whose main objective is to demonstrate experimentally that the following hypothesis holds:

computational developmental systems on a cellular structure are a) naturally fault-tolerant and b) evolvable

. By naturally we mean that the system is not fault-tolerant by explicit design nor due to evolutionary pressure, but rather that the framework insures a high probability of fault-tolerance as an emergent property. In this paper, we propose to study the self-repair capacities of a specific developmental cellular system introduced in [13]. More specifically we compare the toroidal and the non-toroidal cases. Their evolvability is to be presented in details in a further article. All the examples studied here have been evolved to configure an abstract digital circuit. The evolved organisms are subjected to a series of different fault models and their self-repair abilities are reported. From the results exposed here, it can be concluded that, while not systematic, perfect self-repair, and hence fault-tolerance is a highly probable property of these organisms and that many of them even exhibit fully perfect self-repair behaviour under all tests performed.

An evolutionary system is presented which employs an embryogeny model to evolve phenotypes in the form of layout of cells in specific patterns and shapes. It is shown that evolved phenotypes exhibit robustness to damage. How and why these traits appear is discussed and it is conjectured that it is the result of the effects of a complex mapping upon simulated evolution.

This paper demonstrates some examples that show the ability of reaction-diffusion mechanism to code the curvature of forms of multi-cellular systems. The simulation model consists of two layers: the first generates reaction-diffusion waves and the second diffuses chemical substances. The results show that topology changes feedback information to the reaction-diffusion mechanism allowing the control of the morphogenetic process.

In group-living animals, aggregation favours interactions and information exchanges between individuals, and thus allows the emergence of complex collective behaviors. In previous works, a model of a self-enhanced aggregation was deduced from experiments with the cockroach

Blattella germanica

. In the present work, this model was implemented in micro-robots Alice and successfully reproduced the agregation dynamics observed in a group of cockroaches. We showed that this aggregation process, based on a small set of simple behavioral rules of interaction, can be used by the group of robots to select collectively an aggregation site among two identical or different shelters. Moreover, we showed that the aggregation mechanism allows the robots as a group to “estimate” the size of each shelter during the collective decision-making process, a capacity which is not explicitly coded at the individual level.

Using decentralized control structures for robot control can offer a lot of advantages, such as less complexity, better fault tolerance and more flexibility. In this paper the evolution of recurrent artificial neural networks as centralized and decentralized control architectures will be demonstrated. Both designs will be analyzed concerning their structure-function relations and robustness against lesion experiments. As an application, a gravitationally driven robotic system will be introduced. Its task can be allocated to a cooperative behavior of five subsystems. A co-evolutionary strategy for generating five autonomous agents in parallel will be described.

This article presents the first results of a project in underwater modular robotics, called Neubots. The goals of the projects are to explore, following Von Neumann’s ideas, potential mechanisms underlying self-organization and self-replication. We briefly explain the design features of the module units. We then present simulation results of the artificial co-evolution of body structures and neural controllers for locomotion. The neural controllers are inspired from the central pattern generators underlying locomotion in vertebrate animals. They are composed of multiple neural oscillators which are connected together by a specific type of coupling called synaptic spreading. The co-evolution of body and controller leads to interesting robots capable of efficient swimming. Interesting features of the neural controllers include the possibility to modulate the speed of locomotion by varying simple input signals, the robustness against perturbations, and the distributed nature of the controllers which makes them well suited for modular robotics.

Recently, brain models attempt to support cognitive abilities of artificial organisms. Incremental approaches are often employed to support modelling process. The present work introduces a novel computational framework for incremental brain modelling, which aims at enforcing partial components re-usability. A coevolutionary agent-based approach is followed which utilizes properly formulated neural agents to represent brain areas. A collaborative coevolutionary method, with the inherent ability to design cooperative substructures, supports the implementation of partial brain models, and additionally supplies a consistent method to achieve their integration. The implemented models are embedded in a robotic platform to support its behavioral capabilities.

We present an example of the dynamical systems approach to learning and adaptation. Our goal is to explore how both control and learning can be embedded into a single dynamical system, rather than having a separation between controller and learning algorithm. First, we present our adaptive frequency Hopf oscillator, and illustrate how it can learn the frequencies of complex rhythmic input signals. Then, we present a controller based on these adaptive oscillators applied to the control of a simulated 4-degrees-of-freedom spring-mass hopper. By the appropriate design of the couplings between the adaptive oscillators and the mechanical system, the controller adapts to the mechanical properties of the hopper, in particular its resonant frequency. As a result, hopping is initiated and locomotion similar to the bound emerges. Interestingly, efficient locomotion is achieved without explicit inter-limb coupling, i.e. the only effective inter-limb coupling is established via the mechanical system and the environment. Furthermore, the self-organization process leads to forward locomotion which is optimal with respect to the velocity/power ratio.

This paper presents analysis and follow up experiments based on previous work where a neurally controlled simulated agent was evolved to navigate using path integration (PI). Specifically, we focus on one agent, the best one produced, and investigate two interesting features. Firstly, the agent stores its current coordinates in two leaky integrators, whose leakage is partially compensated for by a normalisation mechanism. We use a comparison between four network topologies to test if this normalised leakage mechanism is adaptive for the agent. Secondly, the controller generates efficient searching behaviour in the vicinity of its final goal. We begin an analysis of the dynamical system (DS) responsible for this, starting from a simple three variable system.

This paper is about the design of an artificial neural network to control an autonomous robot that is required to iteratively solve a discrimination task based on time-dependent structures. The “decision making” aspect demands the robot “to decide”, during a sequence of trials, whether or not the type of environment it encounters allows it to reach a light bulb located at the centre of a simulated world. Contrary to other similar studies, in this work the robot employs environmental structures to iteratively make its choice, without previous experience disrupting the functionality of its decision-making mechanisms.

All biological organisms must be able to regulate certain essential internal variables, e.g. core body temperature in mammals, in order to survive. Almost by definition, those that cannot are dead. Changes that result in a mammal being able to tolerate a wider range of core body temperatures make that organism more robust to external perturbations. In this paper we show that when internal variables are regulated via ‘rein control’ mechanisms, decreasing the range of tolerable values increases the area of observed hysteresis but does not decrease the limits of regulation. We present circumstances where increasing the range of tolerable values actually decreases robustness to external perturbation.

This paper explores the performance of a simple model agent using a reactive controller in situations where, from an external perspective, a solution that relies on internal states would seem to be required. In a visually-guided orientation task with sensory inversion and an object discrimination task a study of the instantaneous response properties and time-extended dynamics explain the non-reactive performance. The results question common intuitions about the capabilities of reactive controllers and highlight the significance of the agent’s recent history of interactions with its environment in generating behaviour. This work reinforces the idea that embodied behaviour exhibits properties that cannot be deduced directly from those of the controller by itself.

Motor synergies, i.e. systematic relations between effectors, have been first proposed as a principle in motor control by N. Bernstein in 1935. Thereafter, his idea has inspired many models of motor control in humans and animals. Recently, “linear synergy”, i.e. a linear relation between the torques applied to different joints, was reported to occur in human subjects during directional pointing movements [4]. In this paper, results from experiments in evolutionary robotics are presented to explore the concept of synergies in general and the role of linear synergy in the organisation of movement in particular. A 3D simulated robotic arm is evolved to reach to different target spots on a plane. Linear synergy is not found to be an outcome of the evolutionary search process, but imposing linear synergy as a constraint on artificial evolution dramatically improves evolvability and performance of evolved controllers.

Consider a group of autonomous, mobile robots with the ability to physically connect to one another (self-assemble). The group is said to exhibit

functional self-assembly

if the robots can choose to self-assemble in response to the demands of their task and environment [15]. We present the first robotic controller capable of functional self-assembly implemented on a real robotic platform.

The task we consider requires a group of robots to navigate over an area of unknown terrain towards a target light source. If possible, the robots should navigate to the target independently. If, however, the terrain proves too difficult for a single robot, the robots should self-assemble into a larger group entity and collectively navigate to the target.

We believe this to be one of the most complex tasks carried out to date by a team of physical autonomous robots. We present quantitative results confirming the efficacy of our controller. This puts our robotic system at the cutting edge of autonomous mobile multi-robot research.

A

swarm-bot

is a robotic entity built of several autonomous mobile robots (called

s-bots

) physically connected together. This form of collective robotics exploits robot interactions both at the behavioral and physical levels. The goal of this paper is to analyze the physical performance of a swarm-bot as function of its size (number

n

of s-bots composing it). We present three tasks and the corresponding swarm-bot performances. In all three tasks we show superlinear performances in a range of

n

where the physical forces applied in the structure fit to the robot design. This superlinear performance range helps in understanding which swarm-bot size is optimal for a given task and gives interesting hints for the design of new application-oriented swarm-bots.

Recently, in both the neuroscience and adaptive behaviour communities, there has been growing interest in the interplay of multiple timescales within neural systems. In particular, the phenomenon of neuromodulation has received a great deal of interest within neuroscience and a growing amount of attention within adaptive behaviour research. This interest has been driven by hypotheses and evidence that have linked neuromodulatory chemicals to a wide range of important adaptive processes such as regulation, reconfiguration, and plasticity. Here, we first demonstrate that manipulating timescales can qualitatively alter the dynamics of a simple system of coupled model neurons. We go on to explore this effect in larger systems within the framework employed by Gardner, Ashby and May in their seminal studies of stability in complex networks. On the basis of linear stability analysis, we conclude that, despite evidence that timescale is important for stability, the presence of multiple timescales within a single system has, in general, no appreciable effect on the May-Wigner stability/connectance relationship. Finally we address some of the shortcomings of linear stability analysis and conclude that more sophisticated analytical approaches are required in order to explore the impact of multiple timescales on the temporally extended dynamics of adaptive systems.

Sensing in the dark is a useful but challenging task both for biological agents and robots. Rats and mice use whiskers for the active exploration of their environment. We have built a robot equipped with two active whisker arrays and tested whether they can provide reliable texture information. While it is relatively easy to classify data recorded at a specified distance and angle to the object, it is more challenging to achieve texture discrimination on a mobile robot. We used a standard neural network classifier to show that it is in principle possible to discriminate textures using whisker sensors even under real-world conditions.

This paper investigates evolvability of artificial neural networks within an artificial life environment. Five different structural mutations are investigated, including adaptive evolution, structure duplication, and incremental changes. The total evolvability indicator, E

total

, and the evolvability function through time, are calculated in each instance, in addition to other functional attributes of the system. The results indicate that incremental modifications to networks, and incorporating an adaptive element into the evolution process itself, significantly increases neural network evolvability within open-ended artificial life simulations.

The Species Adaptation Genetic Algorithm (SAGA) was introduced to facilitate the open-ended evolution of artificial systems. The approach enables genotypes to increase in length through appropriate mutation operators and has been successfully exploited in the production of artificial neural networks in particular. Most recently, this has been undertaken within coevolutionary or multi-agent scenarios. This paper uses an abstract model of coevolution to examine the behaviour of SAGA on fitness landscapes which are coupled to those of other evolving entities to varying degrees. Results indicate that the basic dynamics of SAGA remain unchanged but that the rate of genome growth is affected by the degree of coevolutionary interdependence between the entities.

The idea that the biota can regulate the abiotic components of their environment to levels suitable for life has attracted criticism from neo-Darwinian theorists but is still a viable hypothesis. Here we present a model, similar to Daisyworld [1] but more general, which allows for a more extensive study of the compatibility of biotic regulation with evolutionary theory. Results obtained highlight the importance of constraints on the evolutionary process for the emergence of regulation, and set the scene for more comprehensive future study.

This paper proposes a computational model for solving optimisation problems that mimics the principle of evolutionary transitions in individual complexity. More specifically it incorporates mechanisms for the emergence of increasingly complex individuals from the interaction of more simple ones. The biological principles for transition are outlined and mapped onto an evolutionary computation context. The class of binary constraint satisfaction problems is used to illustrate the transition mechanism.

The Baldwin Effect indicates that individually learned behaviours acquired during an organism’s lifetime can influence the evolutionary path taken by a population, without any direct Lamarckian transfer of traits from phenotype to genotype. Several computational studies modelling this effect have included complications that restrict its applicability. Here we present a simplified model that is used to reveal the essential mechanisms and highlight several conceptual issues that have not been clearly defined in prior literature. In particular, we suggest that canalisation and genetic assimilation, often conflated in previous studies, are separate concepts and the former is actually not required for non-heritable phenotypic variation to guide genetic variation. Additionally, learning, often considered to be essential for the Baldwin Effect, can be replaced with a more general phenotypic plasticity model. These simplifications potentially permit the Baldwin Effect to operate in much more general circumstances.

Contrary to indications made by prior researchers, digital logic circuits designed by artificial evolution to perform binary arithmetic tasks can generalise on inputs which were not seen during evolution. This phenomenon is demonstrated experimentally and speculatively explained in terms of the regular structure of binary arithmetic tasks and the nonoptimality of random circuits. This explanation rests on an assumption that evolution is relatively unbiased in its exploration of circuit space. Further experimental data is provided to support the proposed explanation.

Niche construction is the process whereby organisms, through their metabolism, activities, and choices, modify their own and/ or each other’s niches. Our purpose is to clarify the interactions between evolution and niche construction by focusing on non-linear interactions between genetic and environmental factors shared by interacting species. We constructed a new fitness landscape model termed the NKES model by introducing the environmental factors and their interactions with the genetic factors into Kauffman’s NKCS model. The evolutionary experiments were conducted using hill-climbing and niche-constructing processes on this landscape. Results have shown that the average fitness among species strongly depends on the ruggedness of the fitness landscape (

K

) and the degree of the effect of niche construction on genetic factors (

E

). Especially, we observed two different roles of niche construction: moderate perturbations on hill-climbing processes on the rugged landscapes, and the strong constraint which yields the convergence to a stable state.

This paper examines the effect of cultural learning on a population of neural networks. We compare the genotypic and phenotypic diversity of populations employing only population learning and of populations using both population and cultural learning in two types of dynamic environment: one where a single change occurs and one where changes are more frequent. We show that cultural learning is capable of achieving higher fitness levels and maintains a higher level of genotypic and phenotypic diversity.

The concept of modularity appears to be crucial for many questions in the field of Artificial Life research. However, there have not been many quantitative measures for modularity that are both general and viable. In this paper we introduce a measure for modularity based on information theory. Due to the generality of the information theory formalism, this measure can be applied to various problems and models; some connections to other formalisms are presented.

This paper investigates genetic drift in multi-parent genetic algorithms (MPGAs). An exact model based on Markov chains is proposed to formulate the variation of gene frequency. This model identifies the correlation between the adopted number of parents and the mean convergence time. Moreover, it reveals the pairwise equivalence phenomenon in the number of parents and indicates the acceleration of genetic drift in MPGAs. The good fit between theoretical and experimental results further verifies the capability of this model.

An evolutionary reinforcement-learning algorithm, the operation of which was not associated with an optimality condition, was instantiated in an artificial organism. The algorithm caused the organism’s behavior to evolve in response to selection pressure applied by reinforcement from the environment. The resulting behavior was consistent with the well-established quantitative law of effect, which asserts that the time rate of a behavior is a hyperbolic function of the time rate of reinforcement obtained for the behavior. The high-order, steady-state, hyperbolic relationship between behavior and reinforcement exhibited by the artificial organism did not depend on specific qualitative or quantitative features of the evolutionary algorithm, and it described the organism’s behavior significantly better than other, similar, function forms. This evolutionary algorithm is a good candidate for a dynamics of live behavior, and it might be a useful building block for more complex artificial organisms.

In this paper we investigate the evolutionary pressures influencing genome size in artificial organisms. These were designed with three organisation levels (genome, proteome, phenotype) and are submitted to local mutations as well as rearrangements of the genomic structure. Experiments with various per-locus mutation rates show that the genome size always stabilises, although the fitness computation does not penalise genome length. The equilibrium value is closely dependent on the mutational pressure, resulting in a constant genome-wide mutation rate and a constant average impact of rearrangements. Genome size therefore self-adapts to the variation intensity, reflecting a balance between at least two pressures: evolving more and more complex functions with more and more genes, and preserving genome robustness by keeping it small.

We consider models of emergence, adding downward causation to conventional models where causation permeates from low-level elements to high-level behaviour. We describe an architecture and prototype simulation medium for tagging and modelling emergent features in CA-like systems. This is part of ongoing work on engineering emergence.

In this paper, the results of three experiments, in which a genetic algorithm evolves one-dimensional cellular automata (CA), in order to perform the classical main task, are reported. The used systems are not elementary CA but they have a higher number of states. Our aim is to verify if the main-task results are similar to those obtained with elementary CA. Our results confirm that there is a substantial homogeneity.

It is speculated that there is a relationship between 1/

f

noise and computational universality in two-dimensional cellular automata. We use genetic algorithms to find two-dimensional cellular automata which have 1/

f

spectrum. Spectrum is calculated from the evolution of the state of cell from a random initial configuration. The fitness function is constructed in consideration of the spectral similarity to 1/

f

spectrum. The result shows that the rule with the third highest fitness in the experiment has 1/

f

spectrum and it behaves like the Game of Life, although two rules with the highest and the second highest fitness do not have 1/

f

spectrum.

Performing computations with cellular automata, individually or arranged in space or time, opens up new conceptual issues in emergent, artificial life type forms of computation, and opens up the possibility of novel technological advances. Here, a methodology for combining sequences of elementary cellular automata is presented, in order to perform a given computation. The problem at study is the well-known density classification task that consists of determining the most frequent bit in a binary string. The methodology relies on an evolutionary algorithm, together with analyses driven by background knowledge on dynamical behaviour of the rules and their parametric estimates, as well as those associated with the formal behaviour characterisation of the rules involved. The resulting methodology builds upon a previous approach available in the literature, and shows its efficacy by leading to 2 rule combinations already known, and to additional 26, apparently unknown so far.

We compare the long term behaviour of Conway’s Game of Life cellular automaton, from initial random configurations, on a bounded rectangular grid and a bounded Penrose tiling grid. We investigate the lifetime to stability, the final ‘ash’ density, and the number and period of final oscillators. Penrose grids have similar qualitative behaviour but different quantitative behaviour, with shorter lifetimes, lower ash densities, and higher ocurrence of long-period oscillators.

We propose a novel artificial-life-oriented media art “RomperSand”, which applies a three-dimensional version of the Game of Life (GoL) CA for the construction of an interactive virtual playground. In RomperSand, two distinct sets of state-transition rules are combined together: one for simulating physically plausible motion of virtual sand particles and the other for realizing the GoL-like dynamic behavior of

living

structures. Players can operate several virtual tools to create, destroy, and interact with these structures. The system was implemented as a Windows application and was tested by several users, gaining positive appreciations from them.

The investigations carried out about the relationships between the generic dynamic behavior of cellular automata (CA) and their computational abilities have established a very active research area. Evolutionary methods have been used to look for CA with predefined computational abilities; one in particular that has been widely studied is the ability to solve the density classification task (DCT). The majority of these studies are focused on the one-dimensional CA. It has recently been shown that the use of a heuristic guided by parameters that estimate the dynamic behavior of 1D CA can improve the evolutionary search for DCT. The present work shows the application of three parameters previously published in the one-dimensional context generalized to the two-dimensional space: sensitivity, neighborhood dominance and activity propagation were used to evolve CA able to perform the two-dimensional version of the density classification task. The results obtained show that the parameters can effectively help a genetic algorithm in searching for 2D CA. A new rule was found which performed better than others previously published for the 2D DCT.

We study the evolution of simple cells equipped with a genome, a rudimentary gene regulation network at transcription level and two classes of functional genes: motion effectors which allow the cell to move in response to nutrient gradients and nutrient importers required to actually feed from the environment. The model is inspired by the protist

Naegleria gruberi

which can switch between a feeding and dividing amoeboid state and a mobile flagellate state depending on environmental conditions. Simulation results demonstrate how selection in a variable environment affects the gene number and efficiency making the cells to rapidly switch from one expression regime to the other depending on the external conditions.

This paper develops a mathematical model of immune network controlled by cytokines. A software implementation of the model has been applied to intrusion detection in computer network. The obtained results suggest that the performance of the model is unachievable for another approaches of computational intelligence.

High intertidal rocky shores are extremely stressful habitats. Marine snails in these habitats experience highly desiccating conditions, and they locate refuges such as crevices and form dense aggregations of individuals to reduce the effects of desiccation. This study investigates the mechanisms of refuge location in

Melarhaphe neritoides

using a simple set of rules to mimic the behaviour of each individual snail as a computer simulation. Chance interactions with other individuals, other individuals’ trails and the crevices which form part of the virtual environment result in a mainly self-organised pattern of aggregations and crevice occupation which match real patterns obtained in laboratory experiments. Simulations where the following of trails is removed result in a poorer match to the experimental data, indicating the importance of trail-following in establishing these distribution patterns. The study shows that artificial life based models are a potentially useful tool in the investigation of rocky shore systems.

An ant colony shows collective behavior through signal patterns formed by individual ants communicating among themselves. In this paper I devise a method for designing ant colony model and apply the method to design two types of ant colonies, focusing on ant sensitivity to signals. In the first type, I design three foraging models (trail, attraction and desensitization), by modifying a simple foraging model repeatedly, changing ant sensitivity to recruit pheromone to improve foraging by regulating allocation of ants. Out of them, the desensitization model shows the best foraging efficiency as a result of balanced allocation and stable behavior. In the second type, I design a task-allocation model between foraging and mound-piling tasks using independent signals for each task. It shows weak interaction between these tasks.

In this paper, we describe an approach to artificial life, which uses Qualitative Reasoning for the simulation of life within a 3D virtual environment. This system uses qualitative formalisms to describe both the physiology of a virtual creature and its environment. This approach has two main advantages: the possibility of representing integrated physiological functions at various levels of abstraction and the use of a common formalism for the simulation of internal (physiological) and external (environmental) processes. We illustrate this framework by revisiting early work in Artificial Life and providing these virtual life forms with a corresponding physiology, to obtain a complete living organism in virtual worlds.

The transition from unicellular to multicellular organisms is one of the mysteries of evolutionary biology. Individual cells must give up their rights to reproduction and reproduce instead as part of a whole. I review and model the macrocyst stage in slime mould (

Dictyostelium

) evolution to investigate why an organism might have something to gain from joining a collective reproduction strategy. The macrocyst is a reproductive cartel where individual cells aggregate and form a large zygotic cell which then eats the other aggregating cells. The offspring all have the same genetic code. The model is a steady state genetic algorithm at an individual cellular level. An individual’s genetic code determines a threshold above which it will reproduce and a threshold below which it will join a macrocyst. I find that cycles in food availability can play an important role in an organism’s likelihood of joining the macrocyst. The results also demonstrate how the macrocyst may be an important precursor to other cooperative behaviours.

Inspired by the emergent behaviors of ant colonies, we present a novel ant algorithm to tackle unsupervised data clustering problem. This algorithm integrates swarm intelligence and cellular automata, making the clustering procedure simple and fast. It also avoid ants’ longtime idle moving, and show good separation of data classes in clustering visualization. We have applied the algorithm on the standard ant clustering benchmark and we get better results compared with the LF algorithm. Moreover, the experimental results on real world applications report that the algorithm is significantly more efficient than the previous approaches.

We propose a biologically and physically plausible model for ants and pheromones, and show this model to be sufficiently powerful to simulate the computation of arbitrary logic circuits. We thus establish that coherent deterministic and centralized computation can

emerge

from the

collective

behavior of simple distributed markovian processes as those followed by ants.

This paper considers an anisotropic swarm model that consists of a group of mobile autonomous agents with an attraction-repulsion function that can guarantee collision avoidance between agents and a Gaussian-type attractant/repellent nutrient profile. The swarm behavior is a result of a balance between inter-individual interplays as well as the interplays of the swarm individuals (agents) with their environment. It is proved that the members of a reciprocal swarm will aggregate and eventually form a cohesive cluster of finite size. It is shown that the swarm system is completely stable, that is, every solution converges to the equilibrium point set of the system. Moreover, it is also shown that all the swarm individuals will converge to more favorable areas of the Gaussian profile under certain conditions. The results of this paper provide further insight into the effect of the interaction pattern on self-organized motion for a Gaussian-type attractant/repellent nutrient profile in a swarm system.

This paper describes an evolutionary approach to the design of controllers for a team of collective agents. The agents are able to perform ant-like annular sorting, similar to the sorting behaviour seen in the ant species Temnothorax albipennis. While most previous research on ant-like sorting has used hard-wired rules, this study uses neural network controllers designed by artificial evolution. The agents have very simple and purely local sensory capabilities, and can only communicate through stigmergy. Experiments are performed in simulation. The evolved behaviours are presented, analyzed, and compared to previous research on ant-like annular sorting. The results show that artificial evolution is able to create efficient, simple, and scalable controllers able to perform annular sorting of three object types.

This paper considers a group of mobile autonomous agents moving in the space with point mass dynamics. We investigate the dynamic properties of the group for the case that the topology of the neighboring relations between agents varies with time. We introduce a set of switching control laws and show that the desired stable flocking motion can be achieved by using them. The control laws are a combination of attractive/repulsive and alignment forces, and the control law acting on each agent relies on the state information of its neighbors and the external reference signal. By using the control laws, all agent velocities asymptotically approach the desired velocity, collisions are avoided between the agents, and the final tight formation minimizes all agent potentials. Finally, numerical simulations are worked out to further illustrate our theoretical results.

This paper investigates the interaction between cultural evolution and biological evolution in the emergence of phonemic coding in speech. It is observed that our nearest relatives, the primates, use holistic utterances, whereas humans use phonemic utterances. It can therefore be argued that our last common ancestor used holistic utterances and that these must have evolved into phonemic utterances. This involves co-evolution between a repertoire of speech sounds and adaptations to using phonemic speech. The culturally transmitted system of speech sounds influences the fitness of the agents and could conceivably block the transition from holistic to phonemic speech. This paper investigates this transition using a computer model in which agents that can either use holistic or phonemic utterances co-evolve with a lexicon of words. The lexicon is adapted by the speakers to conform to their preferences. It is shown that although the dynamics of the transition are changed, the population still ends up of agents that use phonemic speech.

The complexity, variation, and change of languages make evident the importance of representation and learning in the acquisition and evolution of language. For example, analytic studies of simple language in unstructured populations have shown complex dynamics, depending on the fidelity of language transmission. In this study we extend these analysis of evolutionary dynamics to include grammars inspired by the principles and parameters paradigm. In particular, the space of languages is structured so that some pairs of languages are more similar than others, and mutations tend to change languages to nearby variants. We found that coherence emerges with lower learning fidelity than predicted by earlier work with an unstructured language space.

This paper presents a computational framework for studying the influence of learning on the evolution of avian communication. We conducted computer simulations for exploring the effects of different learning strategies on the evolution of bird song. Experimental results show the genetic assimilation of song repertoires as a consequence of interactions between learning and evolution.

Typically, multi-agent models for studying the evolution of perceptually grounded lexicons assume that agents perceive the same set of objects, and that there is either joint attention, corrective feedback or cross-situational learning. In this paper we address these two assumptions, by introducing a new multi-agent model for the evolution of perceptually grounded lexicons, where agents do not perceive the same set of objects, and where agents receive a cue to focus their attention to objects, thus simulating a Theory of Mind. In addition, we vary the amount of corrective feedback provided to guide learning word-meanings. Results of simulations show that the proposed model is quite robust to the strength of these cues and the amount of feedback received.

One of the greatest challenges in the modern biological and social sciences has been to understand the evolution of altruistic and cooperative behaviors. General outlines of the answer to this puzzle are currently emerging as a result of developments in the evolutionary theories of multilevel selection, cultural group selection, and strong reciprocity. In spite of the progress in theory there is shortage of studies devoted to the connection of theoretical results to the real social systems. This paper presents the model of cooperation which is based on assumptions of heritable markers, constrained resource, and local interactions. Verification of model’s predictions with the real data on aggression in archaic egalitarian societies has demonstrated that initial modeling assumptions are acceptable as major factors of social evolution for these societies.

We study a population of individuals playing the prisoner’s dilemma game. Individual strategies are invariable but the network of relationships between players is allowed to change over time following simple rules based on the players’ degree of satisfaction. In the long run, cooperators tend to cluster together in order to maintain a high average payoff and to protect themselves from exploiting defectors. We investigated both synchronous and asynchronous network dynamics, observing that asynchronous update leads to more stable states, and is more tolerant to various kinds of perturbations in the system.

In human societies, facts are constructed through social consensus. Here, the formation of social institutions in such a society is studied using a multi-agent-based simulation. Institutions are formed through communications among members, and the effects of errors in communication on the formation of institutions are investigated. Our results show that the institution is established when information suppliers frequently make errors in their information interpretation. We propose here that there is a phase transition in the error rate of the information suppliers in the formation of institutions.

Typogenetics was originally devised as a formal system with operations on DNA strands. It was recently demonstrated to be an effective model on which to study the emergence of self-replication by Kvasnicka et al.’s work. We make several extensions and variations on their work. The way of measuring difference and similarity between strands are improved. Many different mappings between doublet codes and their enzyme functions are tried. Triplet codes are also introduced. Through various experiments we observe frequent emergence of autoreplicators. We also find that emergence of self-replicators are robust phenomenon under various environments in typogenetics.

A self-reproducing cycle has the fundamental organization,

$A+X \longrightarrow 2A$

, and is autocatalytic, i.e. the products catalyze the formation of the products. The rate of increase of A is proportional to A, i.e. exponential. Asexual living entities often grow exponentially when resources are abundant, and decay exponentially when resources are scarce, according to autocatalytic kinetics. If two previously independently replicating autocatalytic entities can form a physical union that is still capable of autocatalysis but with a reduced decay rate, then the symbiosis can be viable in an environment in which resources have been depleted, even if the symbiont has a lower growth rate than either of its component particles. A good symbiont possesses the following features: i. low steric hindrance between components, ii. policing of defection or cheating by symbiont components. iii. low decay rate back to components. iv. absence of emergence of active sites susceptible to decay reactions. v. high rate of the final reproductive step. Failure to form stable symbiosis can result from deficits in any of these features, and is a problem central to the origin of both metabolism and template replication.

In this paper, we consider how self-description can be realized for construction and execution in a single framework of a variant of graph rewriting systems, called graph rewriting automata. As an example of self-description for construction, self-replication based on a self-description is shown. A meta-node structure is introduced for self-description for execution that enables us to embed rule sets in the graph structure. It can be a model of systems that maintain and modify themselves.

Recent studies of complex networks offer new methods for characterizing large scale of networks and provide new insights on how such networks are developed. In particular, researchers focused on biological networks such as gene regulatory systems, protein interactions and metabolic pathways in order to understand how these elemental reactions are integrated as an organism. Although various statistical features of network structures, such as scale-free or small-world, have been studied to approach underlying principles of network organization, more detailed analysis on network properties is required to understand their functions.

The community finding algorithm proposed by Girvan and Newman provides another useful technique for investigating topological structures of large networks. Applying this method to metabolic networks, we found that behavior like that of Zipf’s law of the distribution of community size is shared very generally among a wide range of organisms. With the aim of realizing how this property is achieved, we present a new evolutionary model of metabolic reactions based on artificial chemistry.

A Toy Model of an artificial chemistry that treats molecules as graphs was implemented based on a simple Extended Hückel Theory method. Here we describe an extension of the model that models chemical reactions as the result of “collisions”. In order to avoid a possible bias arising from prescribed generic reaction mechanisms, the reactions are simulated in a way that treats the formation and breakage of individual chemical bonds as elementary operations.

Is it possible to create a simple physical system that is capable of replicating itself? Can such a system evolve interesting behaviors, thus allowing it to adapt to a wide range of environments? This paper presents a design for such a replicator constructed exclusively from synthetic DNA. The basis for the replicator is crystal growth: information is stored in the spatial arrangement of monomers and copied from layer to layer by templating. Replication is achieved by fragmentation of crystals, which produces new crystals that carry the same information. Crystal replication avoids intrinsic problems associated with template-directed mechanisms for replication of one-dimensional polymers. A key innovation of our work is that by using programmable DNA tiles as the crystal monomers, we can design crystal growth processes that apply interesting selective pressures to the evolving sequences. While evolution requires that copying occur with high accuracy, we show how to adapt error-correction techniques from algorithmic self-assembly to lower the replication error rate as much as is required.

The classical approach to using utility functions suffers from the drawback of having to design and tweak the functions on a case by case basis. Inspired by examples from the animal kingdom, social sciences and games we propose

empowerment

, a rather universal function, defined as the information-theoretic capacity of an agent’s actuation channel. The concept applies to any sensorimotoric apparatus. Empowerment as a measure reflects the properties of the apparatus as long as they are observable due to the coupling of sensors and actuators via the environment.

Many researchers are developing frameworks inspired by natural, especially biological, systems to solve complex real-world problems. This work extends previous work in the field of biologically inspired computing, proposing an artificial endocrine system for autonomous robot navigation. Having intrinsic self-organizing behaviour, the novel artificial endocrine system can be applied to a wide range of problems, particularly those that involve decision making under changing environmental conditions, such as autonomous robot navigation. This work draws on “embodied cognitive science”, including the study of intelligence, adaptivity, homeostasis, and the dynamic aspects of cognition, in order to help lay down fundamental principles and techniques for a novel approach to more biologically plausible artificial homeostatic systems. Results from using the artificial endocrine system to control a simulated robot are presented.

A framework for natural language processing based on an artificial life model is introduced. Human-computer interfaces require models of dialogue structure that capture the variability and unpredictability within dialogue. In this paper, taking as starting point the notion of eco-grammar system, and by extending it to the concept of Conversational Grammar Systems (CGS), we introduce a new formal framework for conversation modelling.

This paper presents a model of the co-evolution of transmissible disease and a population of non-randomly mixed susceptible agents. The presence of the disease elements is shown to prevent the onset of genetic convergence of the agent population. The epidemiological model also acts in a distributed fashion to counter the tendency of the agent population to occupy spatially close-knit communities. The simulation applies a modified mathematical SIR epidemiological model of disease transmission in combination with the well-studied technique of artificial ecosystems. It includes various aspects of disease transmission that are not usually modelled due to the effort required to incorporate them into mathematical models. These include a distributed agent population with non-uniform infectiousness and immunity as well as a mutable disease model with evolving latency and infections that evolve to prey on diverse agent characteristics.

This paper elaborates upon an idea and a development introduced and presented by Bersini in [1]. Roughly, by observing the search space of a combinatorial problem in a “clever” way, it can be drastically reduced. In order to discover this “clever way”, a second search process has to be engaged in the space of the observables. So two Genetic Algorithms (GAs) are intertwined to solve the whole problem: one in the original space and one in the space of observables of the original one. We are going to present and evaluate this idea on a Cellular Automata (CA) implementation of a binary numbers adder. The experiments show that the new algorithm, combining the two evolutionary searches, speeds up the research and/or increases the quality of the solutions in a significant way.

This research investigates an evolutionary approach to engineering agent collectives that accomplish tasks cooperatively. In general, reproduction and selection form the two cornerstones of evolution and in this paper we study various reproduction schemes in an evolving population of agents. We classify reproduction schemes in temporal and spatial terms, that is, by distinguishing when and where agents reproduce. In terms of the temporal dimension, we tested schemes where agents reproduce only at the end of their lifetime or multiple times during their lifetime. In terms of the spatial dimension we distinguished locally restricted reproduction (agents reproduce only with agents in adjacent positions) and panmictic reproduction (when an agent can reproduce with any other in the environment). This classification leads to four different reproduction schemes, which we compare, via their overall impact upon collective performance. Results using two completely different types of evolvable controllers (hand-coded or neural-net based) indicate that utilizing single reproduction at the end of an agent’s lifetime and locally restricted reproduction afforded the agent collective a significantly higher level of performance in its cooperative task.

After a survey of some typical realizations of self-replicating machines, this paper presents the self-replication based on construction and the self-replication based on inspection of an interactive loop, chosen as an easily understandable example. The construction-based replication process is achieved by translation and transcription of the configuration information of the loop in the processing unit of a data and signals cellular automaton (DSCA). The inspection-based replication process is realized by duplication and translation of the same configuration information in the processing unit of the DSCA.

This paper presents a novel method for coordinating multiple biomimetic robotic fish in box-pushing task. Based on our successfully developing a robotic fish prototype of which the swimming modes can be switched flexibly and smoothly, we step further to study coordination problems of multiple robotic fish in unstructured and dynamic environments. To simplify the difficulties of path planning and action decision when the robotic fish is approaching the box, we employ the

situated-behavior

method, and for each situation a specific behavior is elaborately designed. On dealing with the synchronization and coordinated pushing problems in the particular underwater environment, fuzzy logic method is adopted for motion planning of the fish. Experimental results of box-pushing performed by two robotic fish validate the effectiveness of the proposed method.

Morphogenesis and the spatial structure of an organism have repercussions on gene expression. These effects can influence the results of regulatory network reconstruction. An integrated, flexible and extensible computational framework for modelling gene expression dynamics within spatially growing structures is developed and used as a test system for evaluating a reconstruction algorithm. With complex morphological structures, significant effects of spatial organisation on the reconstruction process are observed. The results also reveal that stronger regulatory interactions result in more frequent cases of indirect regulation, posing a challenge for accurate network reconstruction.

Song communication of artificial birds is simulated in a 2D space, in which male and female birds communicate and then leave their offspring based on their communication performance. The communication is modeled as interaction between different types of finite-state automata, one for song production by males, the other is for song evaluation by females. In addition, an abstract space is introduced for studying how spatial structure affects the evolution of song communication system. We find a correlation between global spatiotemporal patterns and local communications between artificial birds. In particular, we report a habit segregation phenomenon of our simple ecosystem.

In a recent article Chu

et al.

presented a computational model investigating the evolutionary origin of so-called

uptake signal sequences

in bacteria. In that contribution the authors used an agent-based approach. The main aim of this article is to understand the fitness-landscape on which the agents operate. We propose such a fitness- landscape and discuss its implications. This opens the possibility to use GAs for future simulations rather than the computationally expensive agent-based model.

Recently, all the human genes were identified. But understanding the functions coded in the genes is of course a much harder problem. We are used to view DNA as some sort of a computer code, but there are striking differences. For example, by using entropy, it has been shown that the DNA code is much closer to random code than written text, which in turn is less ordered than ordinary computer code. Instead of saying that the DNA is badly written, using common programming standards, we might say that it is written in a different style – an evolutionary style. In this paper the coding style of creatures from the artificial life platform Avida has been studied. Avida creatures that have evolved under different size merit methods and mutation rates have been analysed using the notion of stylistic measures. The analysis has shown that the evolutionary coding style depends on the environment in which the code evolved, and that the choice of size merit method and mutation probabilities affect different stylistic properties of the genome. A better understanding of Avida’s coding style, might eventually lead to insights of evolutionary codes in general.

How can a new incoming biological node measure the degree of nodes already present in a network and thus decide, on the basis of this counting, to preferentially connect with the more connected ones? Although such explicit comparison and choice is quite plausible in the case of man-made networks, like Internet, leading the network to a scale-free topology, it is much harder to conceive for biochemical networks. The computer simulations presented in this article try to respect simple and, as far as possible, basic biological characteristics such as the heterogeneity of biological nodes, the existence of natural hubs, the way nodes bind by mutual affinity, the significance of type-based network as compared with instance-based one and the consequent importance of the nodes concentration to the selection of the partners of the incoming nodes.

Inspired by the phenomenon of symbiosis in natural ecosystem, a master-slave mode is incorporated into Particle Swarm Optimization (PSO), and a Multi-population Cooperative Optimization (MCPSO) is thus presented. In MCPSO, the population consists of one master swarm and several slave swarms. The slave swarms execute PSO (or its variants) independently to maintain the diversity of particles, while the master swarm enhances its particles based on its own knowledge and also the knowledge of the particles in the slave swarms. In the simulation part, several benchmark functions are performed, and the performance of the proposed algorithm is compared to the standard PSO (SPSO) to demonstrate its efficiency.

Efficient hierarchical architectures for reconfigurable and adaptive multi-agent networks require dynamic cluster formation among the set of nodes (agents). In the absence of centralised controllers, this process can be described as self-organisation of dynamic hierarchies, with multiple cluster-heads emerging as a result of inter-agent communications. Decentralised clustering algorithms deployed in multi-agent networks are hard to evaluate precisely for the reason of the diminished predictability brought about by self-organisation. In particular, it is hard to predict when the cluster formation will converge to a stable configuration. This paper proposes and experimentally evaluates a predictor for the convergence time of cluster formation, based on a regularity of the inter-agent communication space as the underlying parameter. The results indicate that the generalised “correlation entropy”

K

2

(a lower bound of Kolmogorov-Sinai entropy) of the volume of the inter-agent communications can be correlated with the time of cluster formation, and can be used as its predictor.

Evolution by natural selection is a process of variation and selection acting on replicating units. These units are often assumed to be individuals, but in a sexual population, the largest reliably-replicated unit on which selection can act is a small section of chromosome – hence, the ‘selfish gene’ model. However, the scale of unit at which variation by spontaneous mutation occurs is different from the scale of unit at which variation by recombination occurs. I suggest that the action of recombinative variation and mutational variation together can enable local optimization to occur at two different scales simultaneously. I adapt a recent model illustrating a benefit of sexual recombination to illustrate conditions for two scales of optimization in natural populations, and show that the operation of natural selection in this scenario cannot be understood by considering either scale alone.

This paper proposes a computational motion control model of a redundant manipulator inspired by biological brain-motor systems. The proposed model consists of two processing layers dubbed “CPG” and “Dynamical memory”. Likewise biological central pattern generators in spinal cord, the CPG layer plays a role in generating torque patterns for realizing periodic motions. On the contrary, the higher brain model, i.e. the Dynamical memory layer is a time-series pattern discriminator implemented by a recurrent neural networks (RNN). By associating time-series of the system states with optimized CPG parameters, the RNN can predictively modulate the generating torque patterns by recalling well-suited CPG parameters according to the sensorimotor time-series.

This paper explores a universal property in the behavior of growing scale-free networks. The characteristic of scale-free networks is that the degree distribution follows the power-law. This structure has been found in various kinds of self-organized networks. Most investigations conducted so far have demonstrated that network topologies are scale-free at a specific point in time. On the other hand, we focus attention on universality in the growing process of networks. In our proposed model, each node has its own fitness to designate the tendency allowing the node to acquire new links. From the simulation results, spread of the network follows the power-law, and power spectrum of the growing process shows 1/

f

noise, not to mention that the network has scale-free structure. It is found that those properties are in common with self-organized criticality. In conclusion, self-organizational growing networks follow the power-law not only in the sense of scale-free characteristic but also in the spatial and temporal sense.

The Synapsing Variable Length Crossover (SVLC) algorithm provides a biologically inspired method for performing meaningful crossover between variable length genomes. In addition to providing a rationale for variable length crossover it also provides a genotypic similarity metric for variable length genomes enabling standard niche formation techniques to be used with variable length genomes. Unlike other variable length crossover techniques which consider genomes to be rigid inflexible arrays and where some or all of the crossover points are randomly selected, the SVLC algorithm considers genomes to be flexible and chooses non-random crossover points based on the common parental sequence similarity. The SVLC Algorithm recurrently “glues” or synapses homogenous genetic sub-sequences together. This is done in such a way that common parental sequences are automatically preserved in the offspring with only the genetic differences being exchanged or removed, independent of the length of such differences. In a variable length test problem the SVLC algorithm is shown to outperform current variable length crossover techniques.The SVLC algorithm is also shown to work in a more realistic robot neural network controller evolution application.

This paper introduces a novel study on the

sense of valency

as a vital process for achieving adaptation in agents through evolution and developmental learning. Unlike previous studies, we hypothesise that behaviour-related information must be underspecified in the genes and that additional mechanisms such as valency modulate final behavioural responses. These processes endow the agent with the ability to adapt to dynamic environments. We have tested this hypothesis with an

ad hoc

designed model, also introduced in this paper. Experiments have been performed in static and dynamic environments to illustrate these effects. The results demonstrate the necessity of valency and of both learning and evolution as complementary processes for adaptation to the environment.