Talks

• Alwyn Barry
  Using Transposons and Flanking Sequences in Evolutionary Computation
  
  A transposon is a mobile block of genetic material that can be moved
  to a different position within a cell; transposons are flanked by a
  sequence of genes known as the flanking sequence.  Simoes and Costa
  (1999a, 1999b, 2000a) proposed three forms of transposition operators
  (asexual, simple and tournament) that could outperform traditional one
  and two point crossover in a genetic algorithm.  Experiments were
  conducted using four test functions and initial results were compared
  to published results.  From these results a number of new
  characteristics defining the behaviour of each transposition operator
  were identified.  Asexual transposition was then applied to another
  evolutionary algorithm, evolution strategies.  Results suggest that
  asexual transposition could not outperform traditional (1+1)-ES.
  However, for an alternative form of asexual transposition, replicative
  asexual transposition, preliminary results are promising.
• Andrew Carnell
  
  Obtaining Precise spike trains in Spiking Neural nets.
  
  Abstract.
  An implementation of STDP is used to modify the synaptic weights of a
  Leaky Integrate and Fire (LIF) neuron, so that the neuron is able to
  produce a precise output spike train in response to a randomly
  generated input pattern consisting of many individual spike trains -
  with each spike train consisting of 5 spikes. It is shown that in order
  to produce accurate output spike trains, it is necessary to have many
  (~500) input spike trains. The STDP training method does not include an
  error feedback mechanism but is still able to produce any randomly
  generated output consisting of 5-10 spikes. It is also shown that a
  single neuron can be trained to respond to multiple, different,
  randomly generated inputs with highly different, yet highly precise
  spiking output responses if trained using an interleaving process in
  order to minimise the forgetting of other I/O associations.
• Pejman Iravani
  
  Collective behaviour study and representation through hyper-network modelling
   
  Understanding collective behaviour is a complex task due to the many interactions and their effects. Network theory can be applied to represent the localised agent interactions, but these are still represented at the lowest description level. Hyper-networks can be applied to represent and model, in a hierarchical manner, agent interactions and their dynamics. This hierarchical organisation allows to discover and encapsulate some of the interesting emerging properties of the system, thus simplifying the understanding of complex systems and their dynamics. 

• Famine relief networks in ants

   

  Dick James

  Department of Physics & Centre for Mathematical Biology, University of Bath.

   

  The use of network theory to probe the social structure of animals is becoming

  more widespread. The next step is to move beyond pattern to process, to look

  at the interplay between the network structure of contacts and the transmission

  of information through a population. I will give a brief account of the first steps

  of a study, involving people working at UWE, Bristol and Bath, to look at

  networks involved in the transmission of food through a colony of ants.

• Matt Jones, Dept. of Physiology and MRC Centre for Synaptic Plasticity,
  University of Bristol
  
  TITLE: Neuronal correlations, phase-locking and coherence during cognition
  and disease
  
  ABSTRACT: Rhythmic neuronal activity is likely to reflect the coordination
  and interactions of networks of neurons during complex cognitive processes
  and behaviour. Accordingly, many psychiatric disorders are associated with
  impaired neural synchrony.
  
  We are using specialized 'tetrode' electrodes to record simultaneously from
  large groups of neurons, measuring synchrony of network activities in
  anatomically and functionally related circuits, both during normal
  behaviour and in models of diseases like schizophrenia and Down syndrome.
• Name: Andreas Krause
  Institutions: University of Bath
  Title: Local Interactions and Case-Based Decisions: An Exploration of
  Learning
  Abstract: We evaluate repeated decisions of individuals using case-based
  decision theory (CBDT), which provides an alternative to rational
  decision-making by relying on the previous experience in similar situations
  to guide the decisions. In many circumstances CBDT can be regarded as a more
  appropriate decision-making model than traditional rational behaviour. In
  our model individuals base their decisions on their own past experience as
  well as the experience of neighboring individuals. Focussing on the learning
  of the best decision, we investigate the importance of how the "good"
  outcome of a decision is determined. Looking at uncorrelated random
  outcomes, correlated random outcomes, minority and majority outcomes, we
  find that while in some instances learning is achieved, in others instances
  it is very slow or even absent. We notice in particular the absence of
  learning in cases where there are no best decisions and the emergence of
  herding in those instances. This is particularly noticeable where
  individuals take into account the experience of a large number of other
  individuals and have a long memory for past experiences.
• Kevin Laland: Animal Social Learning: Problems and Solutions
• o  Speaker's name: Tobias Larsen
  
  o  (optional) coauthors: Rafal Bogacz
  
  
  o  institution:  Department of Computer Science
  
  Bristol University
  
  o  title:  Initiation and Termination of Integration in a Decision Process
  
  
  o  brief (one paragraph) abstract
  
  For integration based decision making, we establish qualitative and
  quantitative differences between three cortical integration models (Race,
  FFI & LCA) in respect to how they are able to initiate an integration
  process without integrating noise prior to stimulus onset, as well as the
  models' ability to terminate the integration after a decision has been
  made to ensure the possibility of subsequent decisions. Our results show
  that the LCA model has advantages over the race model and the FFI model in
  both respects leading to shorter decision times and an effective
  termination process.
• Hagen Lehmann
  
  Social behaviour of most primates, arguably including humans,
  is usually characterised as an continuum reaching from egalitarian to
  despotic. Despotic societies are characterised by a strict hierarchy with
  very few highly intense aggressive interactions, usually unilateral
  directed from dominant to subordinate. The differentiation of ranks of
  individuals in egalitarian societies is less well-defined; aggression is
  frequent but less violent, bilateral,and each individual in such a species
  executes a large repetoire of reconciliation behaviours. In this paper we
  describe our approach to answer the question which selective environmental
  pressures shaped these different social organisations and what is the role
  of dominance in social evolution. Our technique focuses on agent-based
  modelling of the closely related species in the genus {\em macaca}. In our
  model we compared a dominant and an egalitarian species.
  We measured the distribution of dominance values amongst troop members,
  the centrality of dominance and monitored which group had better
  reproduction success under which environmental settings.
  
• LYME: FROM OLD-LYME TO EUROPE
  Can we control a possible disease
  break-out?
  Loukia N. Lili,
  Dep. of Mathematics, Univ. of Bath, UK
  Supervisor: Prof. Nicholas F. Britton, Dep. of Mathematics
  Collaborator: Dr. Klaus Kurtenbach, Dep. of Biology and Biochemistry
  ABSTRACT
  In 1975, Lyme disease was classified as a new disease in response to approxi-
  mately 50 cases of pedriatic arthritis in the town Old Lyme, Connecticut. Since
  then, the disease remains endemic in certain areas of the USA and for the past
  10 years appears as an upcoming threat for Europe. Modelling Lyme disease is
  a challenging task for the current researcher, as it involves a lot of field work, lab
  experiments, analytical and numerical mathematics and excellent interdiscipli-
  nary collaborations. In this presentation, we investigate Lyme disease as a two
  host/two strain vector-borne zoonoses from the mathematical view. The models
  assume that a tick can carry either of the two dominant pathogenic strains in
  Europe and it can be fed either on a bird or on a rodent. Through the ODE
  model, we predict the environmental conditions and the population dynamics of
  a possible disease break-out, whereas through the PDE model, with a diffusion
  effect we looking at how deer distribution (the main reservoir for ticks) affects
  the tick population growth.
• Name:           Andy Lulham
  Coauthors:      Dr. Rafal Bogacz, Prof. Malcolm W. Brown
  Institution:    University of Bristol
  
  Title:
  Anti-Hebbian learning in the perirhinal cortex may underlie both
  familiarity discrimination and feature extraction
  
  Abstract:
  Psychological experiments have shown that the capacity of the brain for
  discriminating visual stimuli as novel or familiar is almost limitless. Recent
  neurobiological studies have established that the perirhinal cortex is
  critically involved in both familiarity discrimination and feature extraction.
  However, opinion is divided as to whether these two processes are performed by
  the same neurons. Two models combining the two processes, based on Hebbian
  learning, have been shown to have a lower capacity than the model specialised in
  familiarity discrimination based on anti-Hebbian learning. It has been proposed
  that the reason for the poor performance of these two combined models is their
  inability to extract independent features. Additionally, anti-Hebbian learning
  is more consistent with experimental data concerning neuronal responses and
  synaptic plasticity in the perirhinal cortex. In this paper we show that a
  well-known model of visual feature extraction, Infomax, which uses anti-Hebbian
  learning can also efficiently perform familiarity discrimination. This model has
  a significantly larger capacity than previously proposed combined models,
  particularly when correlation exists between inputs, as it does in the
  perirhinal cortex. We also discuss further questions that need to be addressed
  to establish if Infomax is a valid candidate for a model of familiarity
  discrimination in the perirhinal cortex.
• Speakers name: Joanna L. Parmley
  Institution: University of Bath
  Title: Splicing and Protein Evolution in Mammals
  Abstract: It is supposed that protein evolution and amino acid content are determined by protein function. Here, I examine whether the requirement to specify information within the exons for the accurate excision of introns might influence both the amino acid content and the rate of protein evolution of mammalian genes. It is shown that the majority of amino acids show skewed usage near intron-exon boundaries. This is shown to be due to selection at the nucleotide level, as there are opposing trends for the 2-fold and 4-fold degenerate codon blocks of arginine and leucine. Most strikingly, those amino acids preferred near boundaries are those with codons enriched in exonic splicing enhancers (ESEs).  The rate of evolution is lowest near intron-exon boundaries, at least in part owing to ESEs, such that domains flanking boundaries evolve on average at approximately half the rate of exon centres from the same gene. This mechanism has more widespread effects, to the extent that the proportion of gene sequence near intron-exon boundaries is one of the strongest predictors of protein evolution in mammals
• Sean A. Rands *1, Rufus A. Johnstone 2, Richard A. Pettifor 3, J. Marcus
  Rowcliffe 3 and Guy Cowlishaw 3
  
  Social foraging and dominance relationships: the effects of socially
  mediated interference
  
  1 - Centre for Behavioural Biology, Department of Clinical Veterinary
  Science, University of Bristol
  2 - Department of Zoology, University of Cambridge
  3 - Institute of Zoology, Zoological Society of London
  
  * Speaker and communicating author: sean.rands@bristol.ac.uk
  
  In socially foraging animals, it is widely acknowledged that the position
  of an individual within the dominance hierarchy of the group has a large
  effect upon its foraging behaviour and energetic intake, where the intake
  of subordinates can be reduced through socially mediated interference. We
  explored the effects of interference upon group dynamics and individual
  behaviour, using a spatially explicit individual-based model. Each
  individual follows a simple behavioural rule based upon its energetic
  reserves and the actions of its neighbours (where the rule is derived from
  game theory models). Dominant individuals have larger energetic reserves
  than their subordinates, and the size of this difference increases when
  either food is scarce, the intensity of interference suffered by the
  subordinates increases, or the distance over which dominant individuals
  affect subordinates increases. These differences in reserves are not based
  upon prior assumptions of the effects of social hierarchy and energetic
  reserves upon predation risk, and emerge through nothing more than a
  reduction in energetic intake by the subordinates when dominants are
  present.
  
  relevant URL: http://seis.bris.ac.uk/~frsar/
• Spike Trains and Mutual Information
  
  Real and artificial spiking neurons communicate by sending each other
  spike trains.  The concept of mutual information MI(X,Y) between spike
  trains X and Y will be described and briefly discussed.  The standard
  method,
  involving bins, for estimating MI(X,Y) is computationally difficult,
  especially for long time intervals.  Another method of detecting and
  quantifying  mutual information will be presented, based on nearest
  neighbour distances in a sample (X1,Y1),..., (XN,YN) of pairs of spike
  trains X and Y.  It is claimed that this is superior to the standard
  method for small samples or long time intervals.
  See www.bath.ac.uk/~masdr/mutual.ps
  
  Daniel Richardson & Carl O'Dwyer
• Name: Tom O. Richardson
  Institution: UWE & UOB
  Title: Evaluation in Teaching Ants
  Abstract: Tandem running is a form of recruitment in ants in which a single
  well-informed worker guides a naïve nestmate to a goal. Tandem running in
  Temnothorax albipennis meets all of the criteria of a strict definition of
  teaching. Here we show that these teaching ants also perform three
  different kinds of evaluation. First, the longer the lesson has proceeded
  the longer the teacher will wait for her pupil to catch up.  Second, ant
  teachers modulate their 'patience' with their pupil depending on the value
  of the lesson, i.e. the value of the goal. Third, ant teachers are more
  patient with quick pupils than slow ones.
• Speaker's name: Elva J H Robinson
  coauthors: FLW Ratnieks, M. Holcombe
  institution : University of Bristol
  title :ANT FORAGING NETWORKS: MODELLING THE NEGATIVE PHEROMONE
  
  brief abstract :Insect societies are complex systems which face the
  challenge of co-ordinating the activities of their many individuals. Ant
  trail pheromones increase system performance by attracting foragers to
  rewarding sections of the colony's trail network. Previous work on social
  insect foraging has focussed on the role of these positive, attractive
  pheromones. Computer simulation studies indicated that the effectiveness of
  foraging trail networks could be greatly improved by repellent or negative
  pheromones (Stickland et al. 1999 in Information Processing in Social
  Insects pp83-100, however, despite the strong theoretical advantage, at
  that time there were no known examples of negative pheromones in ant
  foraging trail systems. Negative pheromone has now been shown in the
  Pharaoh's ant, Monomorium pharaonis (Robinson E. J. H. et al. 2005. Nature
  438, p442). Foragers mark the unrewarding branch at a trail bifurcation
  with a signal which greatly increases the probability of other foragers
  selecting the opposite branch or making a U-turn. Foragers also increase
  their lateral motion on their approach to this signal, showing that the
  signal is volatile and indicating that it may help them locate a rewarding
  branch at a trail bifurcation. However, little is known about the
  interactions of the different pheromones in the ants system. An agent based
  model is used to test hypotheses about the complementary roles of
  attractive and repellent signals. This model predicts that the negative
  pheromone has an important role in conferring flexibility in response to
  environmental change.
  
• Jim Smith: Coevolutionary systems with genes and memes.
  Probably focussing on credit-assignment and fitness issues
• Jane White: Transience and control in ecological systems
  

• o  Speaker's name:        Jiaxiang Zhang

  o  (optional) coauthors:  Dr. Rafal Bogacz

  o  institution:                   Department of Computer Science, University of Bristol

  o  title:                            Optimal decision making with realistic bounds on neuronal activity

  o  brief (one paragraph) abstract:

   

  Making choices among alternatives is an elementary feature of humans and animals. During millions years evolution, because of the pressure of nature selection, the intelligent life is compelled to make optimal decision. Recent neurophysiologic studies suggest that that during perceptual decisions the sensory information supporting alternatives is integrated in certain neuronal populations. Some decision making models have been proposed to describe the neuronal activity and behavior response in such paradigm. However, in these linear models the firing rates of integrator neurons may achieve arbitrarily values, which are biologically unrealistic. In this talk we introduce two types of restrictions on maximum firing rate: reflecting and absorbing boundaries. The boundary mechanism is able to confine the neural activity within certain interval during integration process and hence to make the original models more realistic. We compare the performance of the decision making models with boundaries and present under which condition the model can achieve optimal performance.