| NSC home > group seminars |
Non-Standard Computation Group |
Group seminars list of speakers and abstracts: 2007-08 |
| Speaker | Title and Abstract | Date and Location |
| Simon Poulding | An Efficient Approach to Large-Scale Experimentation on Stochastic Algorithms: Convincing experimentation on stochastic algorithms is often difficult to achieve owing to the inherent variability in the algorithm response. However, when each run of the algorithm takes days to find a solution and requires large-scale computing resources, the task can become extremely challenging. This talk presents an efficient approach to this type of large-scale experimentation using empirical methods drawn from industrial engineering and medical trials. It is illustrated by recent work by members of the SEBASE (Software Engineering by Automated SEarch) project on solving highly-constrained task allocation problems using nature-inspired algorithms. | 1315-1415 19th October, 2007 CS103 |
| Martin Trefzer | Evolution of Transistor Circuits:This work focuses on the analog design automation for FPTA architectures by means of EAs. The advantages of using real h ardware for circuit evolution are the significantly faster evaluation of candidate solutions compared to a simulator and the fact that found solutions are guaranteed to work on a real-world substrate. Genetic operators are presented aiming to facilitate the understanding of evolved FPTA circuits and to find solutions that can be transfered to other technolog ies. A great hope is thereby to possibly discover unusual, new design principles. Furthermore, a MO algorithm is applied , in order to allow for taking the numerous variables into account, that are required for optimizing the topology and th e dimensioning of transistor circuits. The proposed genetic operators and the MO approach are successfully applied to th e evolution of logic gates, comparators, oscillators and operational amplifiers. It is achieved to reduce the resource c onsumption of evolved circuits and in some cases it is possible to generate clear schematics of good solutions. | 1315-1415 2nd Nov, 2007 CS103 |
| Ed Powley | Automorphisms (a.k.a. symmetries) of transition graphs for cellular automata:The transition graph of a cellular automaton (CA) is a directed graph representing the CA's global dynamics. Automorphisms of the transition graph (isomorphisms of the graph onto itself) correspond to symmetries in the CA's global dynamics. Study of the elementary CAs (1-dimensional CAs with two states and neighbourhood radius 1) suggests that the space of CA rules can be partitioned into three distinct classes, according to how the total number of automorphisms varies with the number of cells on which the CA operates. This classification is at least partly related to the qualitative behaviour of the CA itself. For instance, one class contains those elementary CAs capable of producing "chaotic" dynamics from an ordered initial configuration; these CAs exhibit significantly fewer automorphisms than the rest. This suggests a possible relationship between "symmetry" and "complexity" in CAs. | 1315-1415 16th Nov, 2007 CS103 |
| Prof John Robinson | Two types of sex (Using crossover for human face description):I will describe experiments that apply an evolutionary algorithm to human face description. The work was inspired by David MacKay's book Information Theory,Inference and Learning Algorithms . MacKay is generally sceptical about evolutionary algorithms, but he devotes a whole chapter to crossover methods in which he suggests conditions in which they can be effective Chapter 19 I argue that those conditions can be made to apply in face description. Face description is locating human faces in pictures then estimating some of their attributes such as orientation, expression, gender, age and race. Against an existing description method that is based on classical probability modelling, I propose and evaluate an alternative where training is via an evolutionary algorithm. The training process attempts to find the best pixels to use in appearance-based matching for estimation of the different attributes. The matching template is represented as a genome and evolves via both mutation and crossover. This application has some attractive features in that the difference in performance of different parameter values can not only be measured against a clear benchmark but also visualized. I will show a range of results, among which the power of crossover (sex) for determining gender (sex) will be treated in some detail. I am particularly interested in an expert audience's advice about how this work connects with the mainstream of evolutionary algorithms (if at all!). | 1315-1415 30th Nov, 2007 CS103 |
| Dr Steve Smith | An Immune Network Inspired Evolutionary Algorithm for the Diagnosis of Parkinson’s Disease: This talk will present a novel evolutionary algorithm inspired by the protein/substrate binding exploited in enzyme genetic programming (EGP) and artificial immune networks. The immune network inspired evolutionary algorithm has been developed in direct response to an application in clinical neurology, the diagnosis of Parkinson’s Disease. The inspiration for the algorithm and its implementation will be described and its performance on data acquired from Parkinson's disease patients and controls will be considered. | 1315-1415 14th Dec, 2007 CS103 |
|
| ||
| Prof Steve Furber (Manchester) | Bio-inspired computing: the SpiNNaker project: The real-time modeling of large systems of spiking neurons is computationally very demanding in terms of processing power, synaptic weight memory requirements and communication throughput. We propose to build a high-performance computer for this purpose with a multicast communications infrastructure inspired by neurobiology. The core component will be a chip multiprocessor incorporating some tens of small embedded processors, interconnected by an NoC that carries spike events between processors on the same or different chips. The design emphasizes modelling flexibility, power-efficiency, and fault-tolerance, and is intended to yield a general-purpose platform for the real-time simulation of large-scale spiking neural systems. | 1315-1415 18th Jan, 2008 CS103 |
| Dr Marc de Kamps (Leeds) | From attention to combinatorial productivity: the role of a neuronal blackboard architecture: Combinatorial productivity is a hallmark of human cognition: it explains how we can make sense of novel visual scenes or sentences. The key idea is that these are composed in a systematic way of familiar elements or features and that we can decompose them by following the creation process in reverse. An important question is, then, ‘What is the neuronal substrate for this compositional process?’. Recent evidence points to attention as an important prerequisite for compositional relations in vision. I will discuss a theory for the role of attention in a neuronal blackboard architecture formed by visual cortex, discuss experimental evidence for this theory and will argue that a similar neuronal mechanism may play a role in high level cognition. | 1315-1415 1st Feb, 2008 CS103 |
| Paul Andrews | Introducing CoSMoS: Complex Systems Modelling and Simulation infrastructure: CoSMoS is a 4 year EPSRC funded research project aiming to building capacity in generic modelling tools and simulation techniques for complex systems. The project started in October 2007 and involves researchers from Computer Science, Electronics, Chemistry, Architecture, Biology and Immunology from the University of York, University of Kent and University of Abertay Dundee. The development of CoSMoS is case-study driven, and via the process of modelling and simulating various complex systems we hope to develop both a generic modelling and analysis process and massively parallel and distributed simulation environment for complex systems. My talk will focus on what we are trying to achieve and how we plan to go about it. I will show some results from of our initial 5 months of work modelling "case-book" complex systems (ants, boids, L-systems, CAs, scale-free networks etc) and outline what is to come. | 1315-1415 15th Feb, 2008 CS103 |
| Song Zhan | An Artificial Gene Regulation Network Developmental Model for Applications: Biological organisms exhibit characteristics such as robustness, adaptivity, and self-reconstruction that are highly desirable in human designed systems. However the exact mechanisms and origins of such characteristics are still far from understood. Recently, the emerging consensus coming from biological research is pointing to the importance of dynamic networks of gene activity, these have come to be known as Genetic Regulatory Networks (GRNs). The processes focus on gene regulation, and protein synthesis. Macroscopically, every complex multi-cellular organism develops from the zygote cell via the mechanisms of gene regulation and cell signaling, this is generally viewed as development. Inspired by these two mechanisms: GRNs and development, a model is built aimed to capture an abstraction of biological development and identify the essential aspects required to produce small but useful designs. | 1315-1415 29th Feb, 2008 CS103 |
| Patricia Vargas, (Sussex) | Addressing the Evolvability of a novel Non-Classical Artificial Neural Network: Since the definitive establishment of "artificial neural networks" as a promising area of research, an increasing number of models have been proposed. These classical connectionist models were mostly developed taking into account the electrical signaling as the only nature of communication between neurons. However, fairly recent findings have shed light on the real mechanisms of the biological nervous systems by suggesting the existence of chemical signaling by gases that would play the role of neurotransmitters. By drawing inspiration from these latest discoveries, the GasNet model is an attempt to create a novel recurrent artificial neural network, which seeks to combine the electrical and chemical signaling onto a single network. When applied to evolutionary robotics, this particular network model has been proved to express robustness and consistent speed of evolution towards good quality solutions when compared with classical models. It is believed that this would be partially due to a more flexible, loosed coupled spatial embedded architecture, hence corroborating neuronal plasticity. In this talk I will present the original GasNet model and all its derivatives to date. Additionally, I will illustrate that the original GasNet has its evolvability further improved when its neurons are not constrained to an Euclidean space. I will show some preliminary results via simple test-beds and a delayed-response robot task. | 1315-1415 14th March, 2008 CS103 |
|
| ||
| Simon Poulding | The SEBASE project: a practical approach to automated software engineering using search SEBASE is very large project that involves seven researchers at York and a similar number at two other UK universities. Its objectives are ambitious: to take a major move forward in the relatively young field of search-based software engineering (SBSE), in terms of both research and industrial application. In this talk, I will give an introduction to SBSE, discuss the potential impact of the SEBASE project on the field, and describe progress made since the project started 18 months ago. | 1315-1415 2nd May, 2008 CS103 |
| Dr John Woodward (University of Nottingham) |
2 talks for the price of 1:
Complexity in Cartesian Genetic Programming: Cartesian Genetic Programming (CGP) is a relatively new type of representation used in Evolutionary Computation (EC), and differs from the tree based representation in that outputs from previous computations can be reused. This is achieved by representing programs as directed acyclic graphs (DAGs), rather than as trees.Thus computations from subtrees can be reused to reduce the complexity of a function. We prove an analogous result to that of Woodward that the complexity of a function using a Cartesian Program representation is independent of the terminal set (up to an additive constant), provided the different terminal sets can both be simulated. This is essentially due to the fact that if a representation can express Automatic Reused Outputs (Miller 2000), then it can effectively define its own terminals at a constant cost. Occam's razor revisited: Occam's razor states, "given two hypotheses which equally agree with the observed data, choose the simpler". We criticize previous arguments for the validity of Occam's razor. The nature of hypotheses spaces is explored and we observe a correlation between the complexity of a concept yielded by a hypothesis and the frequency with which it is represented when the hypothesis space is uniformly sampled. We argue there is not a single best hypothesis but a set of hypotheses which are semantically equivalent, whereas Occam's razor suggests. | 1315-1415 16th May, 2008 P/T/005 |
| Nick Owens (Electronics) | Modelling Biology with Process Algebras to build Biologically Inspired Algorithms. :Biological systems seem to display highly robust remarkable information processing capabilities, often these capabilities are found lacking in human engineered systems. A biological-inspired algorithm attempts to capture the features of the biology and replicate them in a engineered, artificial, enivornment. However creating an algorithm which successfully reproduces the desired properties of biology is a very difficult task. This talk presents the application of process algebra techniques, predominantly related to the Stochastic Pi-Calculus, which help to both understand biology and aid the transition from biology to algorithms. The techniques are demonstrated with examples from a model of the tunablility of signalling pathways in the T lymphocyte of the immune system, that will be used to build new immune inspired algorithms. | 1315-1415 30th May, 2008 CS103 |
| Dr Julian Miller | Searching for the genes for learning: evolving neuro-inspired developmental programs: Although artificial neural networks have taken their inspiration from natural neurological systems, they have largely ignored many aspects that have been revealed by developments in neuroscience. Indeed, evolutionary approaches have mainly assumed that neural learning is associated with the adjustment of synaptic weights. In this talk, we examine a new evolutionary approach to finding the computational functions for learning that are analogous to sub-components of biological neurons. Using a form of genetic programming called Cartesian Genetic Programming, we have devised a compartmental model of a neuron which is a collection of seven chromosomes encoding distinct computational functions. The model allows neurons, dendrites, and axon branches to grow, change or die while solving a computational problem. Starting from a small random network of soma, dendrites, and neurites that develops during problem solving, we have evaluated the learning potential of this approach in a competitive co-evolutionary learning environment. This model has been evaluated on two classic AI problems, Wumpus World and Checkers, with promising results. The ultimate aim of this work is to obtain programs that build neural networks that can self-learn through experience. | 1315-1415 13th June, 2008 CS103 |