POSTGRADUATE SEMINARS NOTE: Seminars in this series prior to Spring 2004 are listed on a separate archive page. Visit http://events.cs.bham.ac.uk/seminar-archive/research for more information. -------------------------------- Date and time: Wednesday 27th August 2003 at 13:00 Location: UG40, School of Computer Science Title: GA or GP, that is NOT the question Speaker: John Woodward (http://www.cs.bham.ac.uk/~jrw) Abstract: Genetic Algorithms (GAs) and Genetic Programming (GP) are often considered as seperate but related fields. Typically, GAs use a fixed length linear representation, whereas GP uses a variable size tree representation. This talk argues that the differences are unimportant. Firstly, variable length actually means variable length up to some fixed limit, so can really be considered as fixed length. Secondly, the representations and genetic operators of GA and GP appear different, however ultimately it is bit strings in the computers memory which is being manipulated whether it is GA or GP which is being run on the computer. The important difference lies in the interpretation of the representation; if there is a one to one mapping between the description of an object and the object itself (as is the case with the representation of numbers), or a many to one (non-uniform) mapping (as is the case with the representation of programs). This has ramifications for the validity of the No Free Lunch theorem, which is valid in the first case but not in the second. It is argued that due to the highly related nature of GAs and GP, that many of the empirical results discovered in one field will apply to the other field, for example maintaining high diversity in a population to improve performance. -------------------------------- Date and time: Monday 22nd September 2003 at 13:00 Location: UG40, School of Computer Science Title: A constant complexity algorithm for solving the Boolean Satisfiability Problem on quantum computers Speaker: Ahmed Younes Abstract: Quantum parallelism is considered as the magic key, which gives the quantum computers the ability to do some types of computation more powerfully than any classical computer, where processing can be done simultaneously on exponential number of states. In this talk we will explain the concept of quantum parallelism and how it can be used to solve the Satisfiability Problem; which is a famous NP-complete problems, on quantum computers. The algorithm runs is a constant number of steps; _O(4)_ , with any given number _n_ of Boolean variables. We will show that in contrast to classical algorithms that the ability of the algorithm to solve the problem increases as the number of variables increases. -------------------------------- Date and time: Monday 1st December 2003 at 13:00 Location: UG40, School of Computer Science Title: Bluetooth and its rendezvous layer Speaker: Marie Duflot (http://www.cs.bham.ac.uk/~mxd) Abstract: I will give a presentation about the Bluetooth protocol, focusing mainly on the rendezvous layer, i.e. the way in which devices establish contact with each other. If you are a specialist in Bluetooth or other wireless comunication protocols, this talk will be quite basic for you (but you may help me to answer some questions :o) ). I'm new to "real-life" protocols in general, and I'm studying Bluetooth in order to verify some properties related to "quality of service", using the two model-checkers UPPAAL and PRISM. -------------------------------- Date and time: Monday 1st March 2004 at 13:00 Location: LG34, Learning Centre Title: Modelling and checking infinite state systems using approximation methods Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Abstract: I will present a method for model checking a large family of infinite state Markov Decision Process. It is based on an approach using "approximations" that I have been developing. The (very simple) idea is that is is not necessary to have a complete knowledge of a system to answer (with absolute certainty) questions about this system. The questions are the classical PCTL formulas. (PCTL = Probabilistic Computation Tree Logic) I am also interested in building abstractions of systems, and will present a mathematical framework for thinking about them. Note about this talk: Anyone interested in model checking, but not familiar with PCTL's syntax or semantics, is invited to let me know. -------------------------------- Date and time: Monday 8th March 2004 at 13:00 Location: LG34, Learning Centre Title: Life in Silico Speaker: Dominique Chu Abstract: One of the aims of Artificial Life research is to create a living entity in silico. It is generally thought that such a simulated entity should possess a membrane and a metabolism, and be able to repair itself, replicate and evolve. In this talk, I will discuss some theoretical and experimental issues associated with the creation of life in silico. -------------------------------- Date and time: Monday 15th March 2004 at 13:00 Location: LG34, Learning Centre Title: Simulating Infant-Carer Relationship Dynamics Speaker: Dean Petters (http://www.cs.bham.ac.uk/~ddp/) Abstract: Advances in autonomous agent technology have resulted in the potential for implementations of multiple agents to act as psychological theories of complex social and affective phenomena. Simulating attachment behaviours in infancy provides a relatively simple starting point for this type of theory development. The presence of neurophysiological, psychological and other types of data facilitates the validation of architectural theories by constraining these architectures at multiple levels. A seven part design process is described which details how requirements are specified and how design, implementation and evaluation processes are carried out. Two competing theories are proposed, one that involves some deliberation and one that is reactive only. -------------------------------- Date and time: Monday 29th March 2004 at 13:00 Location: LG34, Learning Centre Title: PErla Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Abstract: ... and what about having an easy language to describe networks and, more generally, mobile systems? ... and what about using PRISM to do the model checking without knowing the language of PRISM? ... and what if this language was in fact a programming language? If one of these questions seems interesting to you, or if you simply like cake, you are most welcome (^.^) ---- PErla is a (powerful) language that I have designed for modelling network-like systems, and more generally mobile systems. It is a programing language which can be used to express mobile systems' behaviours, entirely decided and easily described by the user. PErla comme with a program which can automatically translate a PErla program into a PRISM input file. This allow any user of PErla to concentrate only on the system behaviour and not on the model checker input file syntax (namely PRISM's language). This talk will be substantially less mathematical than my previous cake talk, and able to interest any "model checking" person, but also people interested in describing systems or designing languages :-) Pascal -------------------------------- Date and time: Monday 14th June 2004 at 13:00 Location: LG 53 - Learning Center Title: Distributed Simulation of Agent-Based Models Speaker: Robert Minson (http://www.cs.bham.ac.uk/~rzm) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: Very large or complex simulation models (such as cognitive AI systems), can eat-up computational resources at a very high-rate, both in terms of memory and processor cycles. One solution to this is distributed simulation, where a large model is split up over many nodes with each node taking responsibility for computing the behaviour of one or more entities, who then interact through a virtual environment distributed across the network. The downside of such systems is that the parallel logic introduced greatly complicates the development process, when the modeller should be thinking about the their model. Conversely, Agent-Toolkits (such as Aaron's SIM_AGENT system) aim to remove such extraneous boiler-plating from the development process. By providing infrastructure for activities such as event-systems, message passing and scheduling, they allow the modeller to focus on the important stuff. But we still have the problem of model-scalability. Over the last 9 months we have created middleware that maps from the model-design toolkit RePast to the (ostentatiously named) distributed simulation framework 'The High Level Architecture' (HLA). I will talk about the design issues for such systems, look at techniques we have used and discuss the level of success, both in terms of performance gains for specific models and in terms of the transparency achieved in the mapping. -------------------------------- Date and time: Tuesday 22nd June 2004 at 13:00 Location: UG40, School of Computer Science Title: Function Set Independent Genetic Programming. Speaker: John Woodward (http://www.cs.bham.ac.uk/~jrw) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: The choice of function set can significantly affect the performance of a Genetic Programming run, possibly due to the difference in complexity of solutions when expressed under different function sets. If modularity is permitted in the representation, the complexity of a solution is independent of the function set (up to a constant). This result motivates a representation which defines a landscape that can be searched independently of the function set. Modules are represented as look up tables (which provides the definition of the module), which are later implemented in the chosen function set after a suitable decompostion has been found. This algorithm also uses an incremental approach to testing. Instead of having a fixed test set for the duration of the evolution, the number of test cases in the test set is gradually incremented as the population learns to solve the current test set. A number of observations are made about this approach. -------------------------------- Date and time: Monday 28th June 2004 at 13:00 Location: UG40, School of Computer Science Title: Bulk loading the M-tree for query performance Speaker: Richard Swinbank (http://www.cs.bham.ac.uk/~rjs) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: The M-tree is a paged, dynamically balanced metric access method that responds gracefully to the insertion of new objects. Like many spatial access methods, the M-tree's performance is largely dependent on the degree of overlap between spatial regions represented by nodes in the tree, and minimisation of overlap is key to many of the design features of the M-tree and related structures. We present a novel approach to overlap minimisation using a new bulk loading algorithm, resulting in a query cost saving of between 25% and 40% for non-uniform data. The structural basis of the new algorithm suggests a way to modify the M-tree to produce a variant which we call the SM-tree. The SM-tree has the same query performance after bulk loading as the M-tree, but further supports efficient object deletion while maintaining the usual balance and occupancy constraints. -------------------------------- Date and time: Monday 2nd August 2004 at 13:00 Location: UG40, School of Computer Science (To be confirmed) Title: Modelling probabilistic mobility with PRISM (Part 1) Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: I will present two talks, the first one on the 2nd August and second one two weeks after (16th August) about some "blue sky thoughts" related to the definition of a mathematical framework, the design of a language able to model (non trivial) probabilistic mobile systems (mostly networks) and the issues encountered when implementing this language/systems in PRISM. Actually these thoughts are not that "blue" (or that "sky" whatever you want it) as this work is the content of a paper I am writing in this moment. I hope to receive some interesting suggestions from the audience. I will concentrate myself on the mathematical framework and language design in the first talk and on the implementation in the second. The first talk will be full of MDPs and the second full of variables :-) Everyone is welcome (but above all PRISM related people). Pascal. -------------------------------- Date and time: Monday 9th August 2004 at 13:00 Location: UG40, School of Computer Science Title: Coordinated planning vs distributed CSP Speaker: Dave Gurnell (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: Autonomous planning agents need to be able to communicate with each other to prevent conflicts between their plans. "Coordinated planning" refers to planning by members of a team of non-competitive agents. The objective of each agent is to produce a complete plan that achieves its own goals whilst remaining coordinated with the plans of the other agents in the team. In this talk I will discuss a new approach to coordinated planning in a Hierarchical Task Network (HTN) domain. In particular I will discuss my current quest for a suitable distributed search algorithm, and the ideas I hope to borrow from well-established fields such as distributed constraint satisfaction. -------------------------------- Date and time: Tuesday 10th August 2004 at 13:00 Location: UG40, School of Computer Science (To be confirmed) Title: 1.5 : Mobility, MDPs, Languages, Semantic, PRISM, Topology, Nodes, Agents, Program, pre-scheduling, Compiler, and more.... Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: In this talk I will present the updating of the language I presented last Monday following the very useful suggestions given by the audience. Not the unusual day of the week following the fact that Dave will be giving his talk at the regular slot (Monday 9th, 1 pm), that I took the week after for the part 2, and that for some reasons I don't want to use 23rd or 30th August. Ok, if you don't follow, see the web page : http://www3.cs.bham.ac.uk/seminars/seminar_details.html?seminar_id=124 (^.^) Pascal. ps: If you came last time last time and plan to come again , the following might be of some interest for you. [1] The language I presented last time is now the low level language of the story. I removed the very bad idea of pointers, and simplified the global shape of terms. I removed the command which assign the content of the (first entry of the) mailbox to a variable (which was difficult to well define) but will show how the same can be done in another way. The "main" language is now a C-like language very more natural to use. The exact grammar is given in BNF by (where c is a basic command). m = c; = c;m = (g){m} :: (g){m} :: ... :: (g) {m}; = [p]{m} :: [p]{m} :: ... :: [p]{m}; = while(g){m} = if(g){m}else{m} The third line expresses a non deterministic choice of terms and the fourth a probabilistic choice. nb: I was too lazy to define a native "for" loop, but this is doable :-) With this grammar an endless random generator of bits which broadcast its bits over the network could be written while(true){ [0.5] {x:=0;} :: [0.5] {x:=1;}; broadcast(x); } I know, I know, I can do that in PRISM without my language, but see that as a "Hello world !" example. [2] I will present the algorithm of the translation between the two languages (from high to low) (Low to PRISM will be done in the next talk). [3] The language and related MDPs, can now define agents, that can move from one locality to another one. This needed an extension of the set of basic commands, but not difficult to do. The mathematical framework remains more or less the same. Agents can act on the node defined at their localities and also on the (very new) locality "shared environment". More details in the talk.... -------------------------------- Date and time: Monday 23rd August 2004 at 13:00 Location: UG40, School of Computer Science Title: Is this a Grand Challenge? Speaker: Marcos Quintana (http://www.cs.bham.ac.uk/~miq) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: This is an unmissable cake talk. Perhaps you'll remember this day as the one that lead you to solve that unsolvable problem and that awarded you the Nobel prize, perhaps you'll remember it as that funny day when a naive Research Student believed he discovered the wheel. Who knows?, but why not give it a try? I will talk about big names in Science such as David Hilbert, John Nash and Amartya Sen. I will talk about games theory and a new way to see them. I will talk about a new play ground where you can simulate and apply your deepest loved computational or AI technique and prove is the (approximately) best to partially solve the problem. I believe this new way to see a set of problems is a Grand Challenge, I believe it may have serious implications to solve important problems for humanity and I believe it will capture the imagination of the general public. I look forward to see you attending, particularly if you are a Professor or you are involved with the Grand Challenges in Computer Science. -------------------------------- Date and time: Friday 3rd September 2004 at 13:00 Location: UG40, School of Computer Science Title: Embryological Modelling of the Evolution of Neural Architectures Speaker: Chris Bowers (http://www.cs.bham.ac.uk/~cpb) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: The architecture of the human brain, and in particular the existence of functionally specialized neural modules, is becoming increasingly well mapped out. However, some of the early neural computational modelling in this area [1] has been thrown into doubt by more detailed evolutionary simulations [2]. There remain a range of feasible computational approaches to understanding neural architecture [3], and we believe the best way forward is to look at the evolution of these structures at a much lower level of description than is often used. Embryogeny is the process by which an embryo structure develops from an initial stem state. This genotyping to phenotype mapping process is adaptive in the sense that the rules of growth are dependent upon the state of the system and vice-versa, forming a dynamical system [4]. The modelling of such processes in a simulated evolutionary framework is known as computational embryogeny. We shall present work where such an approach is utilised to grow neural architectures from an initial stem state based upon a simple model encapsulating individual cell states and diffusion of chemicals in a 3-dimensional space. This computational embryogeny is sufficiently expressive to account for features such as topology and connectivity, and the parameters that specify the initial weights and network learning. Results will be presented from a systematic series of computational experiments which explore the possibility of using this approach to study the emergence of structures appropriate for the simplified 'what-where' task used previously [1,2]. In particular, we consider the impact that different selection pressures towards faster growth and learning might have upon the evolved neural architectures. For completeness, the resulting structures are then compared against a series of more abstract models of the type studied previously [2], but with a matched evolutionary regime. In this way we can gain a better understanding of the reliability of, and relationships between, models at different levels of description. -------------------------------- Date and time: Tuesday 5th October 2004 at 12:00 Location: UG40, School of Computer Science Title: How fast does a network grow ? Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: I am working on modeling (for model checking purposes) dynamic networks. Basically networks in which nodes host several programs, can move to different locations and do others interesting things. In particular (nodes) can spawn new nodes and new programs (in the full version of the modeling they can also kill existing components...). A particular mathematical statement involving PCTL formulas features (two) real numbers that I haven't so far been able to compute, but if I could only bound them (efficiently) the statement remains true and above all becomes useful. These two numbers are involved in the possibility to decide a PCTL formula in the infinite system using only a finite approximation. So the talk will be about how to measure the "entropy" of a dynamic network by the only syntactic analysis of the set of program, where "entropy" is to be understood as the speed of evolution of the size of the system; (smaller is this entropy, easier is the model checking of the infinite state system). This work is the following of what I put in my report 4, but I will not assume (of course) that it has been read and will introduce all notions needed to understand what is going on. The main reason why I am interested in giving this talk is that very surprisingly the definition of the entropy of a program (given the entropy of the basic commands of the language) uses a convolution of distributions. I would have never thought that studying programs would involve such beautiful mathematical formulas :-) Note that this talk is open to everyone, there will be (almost) no model checking stuff and might interest everyone who works on distributed and dynamic systems. I will also talk about an extension of the problem that I have not solved (yet). (^.^) Pascal. -------------------------------- Date and time: Tuesday 19th October 2004 at 12:00 Location: UG40, School of Computer Science Title: alpha-systems are e-approachable Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: This talk is the following of my previous one. Now that the entropy is fully defined (needed to be redefined from scratch but the idea remains the same). I will present the main result of this work which states that if the set of programs of the system have got a certain property then the (infinite state) system is e-approachable (and then model checking in this system is decidable using standard methods). The difficult/interesting part was to set the "weakest" condition on the set of programs which ensure the e-approachability of the system. Another time the most natural way seems to be the good one :-) I will take time at the end to compare this approach to, for instance, the one taken by Panangaden in recent works. Basically the main difference is that my programs/processes have a lot of power on the system, in particular can spawn new components. Pascal. -------------------------------- Date and time: Tuesday 15th March 2005 at 12:00 Location: UG40, School of Computer Science Title: A Structural Operational Semantics for an Extention of KLAIM Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: Klaim*, is an entention of the well known KLAIM (Kernel Language for Agent Interaction and Mobility) that I have developped. Intially started as an attempt of a small extension of the language, this work turned out to be the opportunity to answer the following question : "it is possible to provide a complete SOS semantics for a language which features simultaneously classical control structures, nondeterminism, propabilistic choices and error handling in an environment involving communication failures". The answer is yes, and I will also show an MDP-based operational semantics (leading to model checking). I think it is the first time ever that such a language is proposed (with an actual working SOS semantics). The modelling power will be illustrated with an implementation of the IPv4 zeroconf protocol. Pascal. -------------------------------- Date and time: Tuesday 22nd March 2005 at 13:00 Location: UG40, School of Computer Science Title: Mobility in Reactive Modules, or when KLAIM* meets PRISM Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: Following my previous talk in which I introduced KLAIM* (an extension of KLAIM), I will show the principles of an automatic translation of (a small restriction of) KLAIM* networks as PRISM modules. This translation takes the form of an abstractly defined automatic compiler that I will present. It seem that this method could be used with other process algebra (for instance the pi-calculus), or other kind of "complex" systems but I haven't got time to look at them in details. Pascal. -------------------------------- Date and time: Thursday 31st March 2005 at 15:00 Location: LG34, Learning Center (opposite SoCS) Title: Building agents to understand infant attachment behaviour Speaker: Dean Petters (http://www.cs.bham.ac.uk/~ddp/) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: This paper reports on autonomous agent simulations of infant attachment behaviour. The behaviours simulated have been observed in home environmentsand in a controlled laboratory procedure called Strange Situation Experiment.The Avoidant, Secure and Ambivalent styles of behaviour seen in these studies are outlined, and then abstracted to their core elements to act as a specification of requirements for the simulation. A reactive agent architecture demonstrates that these patterns of behaviour can be learnt from reinforcement signals without recourse to deliberative mechanisms. A number of other competing architectural designs are proposed to support the same range of observed behaviour. These designs are all variants of a general hybrid design, and set out intermediate stages between reactive and fully deliberative architectures. This paper is to be submitted to a IJCAI05 workshop on Modeling Natural Action Selection; deadline after the talk; hence a very tightly scheduled caketalk. I'd welcome comments of the draft of this paper which can reached via my homepage and directly from: http://www.cs.bham.ac.uk/~ddp/att_ijcai05.pdf -------------------------------- Date and time: Tuesday 26th April 2005 at 13:00 Location: UG40, School of Computer Science Title: Towards a modal logic for Klaim* Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: I am investigating a modal logic for my language Klaim* and will present something which can be seen as a mix between standard (non probabilistic) modal logic and the well known Probabilistic Computation Tree Logic. The main idea here is to be able to express properties of elements of a Klaim* network in the more natural way (using non probabilistic modal formulas) but in a situation which can be described by (heavy) use of MDPs. I would like to share my thoughts to 1. see how (natural or unnatural) will my approach be judged by the audience. 2. make my ideas clear about it. (^.^) Pascal. -------------------------------- Date and time: Tuesday 24th May 2005 at 13:00 Location: UG40, School of Computer Science Title: Mobility in Multi-Layered Universes Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: The main reason for this work is that there exists (up to my knowledge) no language or framework witch allows to express the fact that a communicating (and possibly moving) entity may be on several networks at the same time. By "several networks" I really meant, at least in the early stages of this work, a laptop wired-plugged in a Local Area Network and moving in a Wireless Local Area Network. The rules of communications of those two networks are quite different. This difference is illustrated by the frameworks used to model them separately. In the case of a LAN one would rather go toward a graph-based description, and in the case of a WLAN, something else, closer to the standard 3D description of space; the latter in order to take account of the fact that the strength (and then the quality) of communication signals may change with the relative position of the two objects. ("Position" then have two completely different meanings according to the network on which one focuses) Thinking of two layers (LAN and WLAN) and ending up with $n$ layers was a temptation that I felt in. This is just the kind of generalization that some people like falling in, in order (this is at least what they claim) to see things in a "more powerful and simple way". In the present case, this generalization has actually come after having solved a preliminary problem which is an important aspect of the work: what have a LAN and a WLAN in common ? In there any way to see them as two instances of the same object ? Then another problem question came. When someone works with web-agents (for instance) the motion of such computational entities maybe seen as the result of their internal computation. But in the case of my laptop brought by me in some remote places of the School, it would not be suitable to think that its exact position is the result of its internal computation, but rather the result of an external force (in this case the decisions of my mind of going here and not there...). Such motion is to be taken account since the exact position of the laptop (related to quality of the communication which the nearest base station) does matter for its future possible internal states. The question is then: how to express such motion ? Things then start to be a bit complicated when one do not assume that only one computational entities may occur in the environment (while talking to preston using a CAT5 plug on the floor and while streaming some music on the WLAN, my laptop might also be happy to chat with my bluetooth mobile phone....). Our semantics rules must be independent of the exact nature of the possible object that could pop up in the environment. This choice of generalization has been difficult to handle, but afterwards, has shown something important: communication and mobility in Multi-Layered Universes require only 3 rules. This last result is exiting enough to be presented. -------------------------------- Date and time: Tuesday 7th June 2005 at 13:00 Location: UG40, School of Computer Science Title: Attention as a requirement for an embodied cognitive agent. Speaker: Nick Hawes (http://www.cs.bham.ac.uk/~nah) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: The title may seem a little serious for a cake talk, but it is my intention to keep the discussion at a relatively high level. I will start the talk by throwing out the word 'attention' due to its common-usage baggage, and then redefine the title of the talk in terms of 'selection' (i.e. selection from a range of information and processes). With this out of the way I will briefly present an architecture for (linguistic) communication developed by our CoSy partners at DFKI, and then discuss a couple of ways in which it can be extended to incorporate 'selection'. I will then present a more detailed design for one of these extensions: a system for selecting which regions of an image should be preferred when doing visual search (i.e. visual attention). This approach is based on existing work by various authors (*) and includes an approach to learning contextual knowledge using an auto-associative network (which is still in the early stages of development). (*) e.g. http://dx.doi.org/10.1016/j.visres.2004.07.042 Familiarity with page 7 of the current Buzz is not a prerequisite of attendance. Baked goods will be provided, probably in the form of flapjacks. -------------------------------- Date and time: Monday 17th October 2005 at 13:00 Location: UG40, School of Computer Science Title: Ink and Prelude (a live demo) Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: This talk will be a live demo of two modeling languages that I invented recently: Prelude and Ink. Prelude is a high level multi purposes modeling language. It is based on the idea that one should be able to describe objects in the most natural way and write rules to indicates how do they interact to each other. In fact writing Prelude-rules is very similar to write the semantics of process-algebras-based languages as Prelude rules are implicitly universally quantified and interpreted as such. Therefore Prelude files are short, focus to the essential, and one do not have to take care of the encoding of objects internal variables. Prelude files are automatically translated into PRISM files, using a translator that I have written. The translator handles itself the encoding of Prelude objects and the translation of Prelude rules as PRISM commands. Ink is a modeling language, similar to KLAIM* (the variant of KLAIM -- Kernel Language for Agents Interaction and Mobility-- that I have written some times ago). It features processes, able to move between locations and able to communicate with remote locations by use of a system of tuple spaces. Using another translator (that I also have written -- which was a bit more complex than the first one), Ink files can be automatically translated into Prelude files. Ink- processes and Ink-tuples are encoded as Prelude-objects, and the (somehow) complex interaction between processes and tuples spaces (which are nothing else than the translation of the semantics of the language) are encoded as Prelude rules. And then you should ask "... so you have a complete system to probabilistically model-check mobility featured process algebras ?" and I would reply "... something like that." The formal grammars of the two languages as well as the programs can be found on my School web page. You can have a play with them before (or after) the talk (please report any bugs, but I doubt that there are any...). You should be familiar with my work on KLAIM* to fully understand the meaning of Ink languages elements, but some relevant documentation can also be found on my web page. The complete formal semantics of the two languages will be available soon (just haven't got time to finish to write the latex file). In many ways, this talk (at least the design of Ink), can be seen as the natural follow up of another cake talk called "Model-Checking of Complex Structures" that I gave earlier this year. The main ideas behind Prelude (and its translation) come from some that I presented in my last cake talk called "Mobility in Multi-Layered Universes". -------------------------------- Date and time: Monday 24th October 2005 at 13:00 Location: UG40, School of Computer Science Title: Introducing the Cake Talk Challenge: Predictive Text Speaker: Nick Hawes & David Brooks (http://www.cs.bham.ac.uk/~nah) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: In this talk we will discuss our idea of using the Cake Talk series to encourage basic collaborative research within the department using ongoing challenges. The idea is that a group of people propose a fairly general problem, and then researchers from different fields apply various approaches to solving it. Following this introduction, we will describe the first of these challenges: investigating ways of improving the T9 predictive text entry system. We will give an extremely high-level overview of the problem, including an introduction to T9 itself, and then describe one existing solution based on a network of common-sense knowledge. We will conclude the talk with a discussion of other possible approaches to the problem (hopefully based on audience suggestions), and outline our future plans for this challenge (including making a software toolkit available for the development of other approaches). The talk will be informal and high-level (whilst hinting at the technical complexities of the problem), and discussion will be actively encouraged. -------------------------------- Date and time: Monday 31st October 2005 at 13:00 Location: UG40, School of Computer Science Title: Specification and compilation into runtime monitors, of high-level security policies for untrusted programs Speaker: Andy Brown (http://www.cs.bham.ac.uk/~ajb) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: Any program executed by a host computer that has previously been transmitted across a network is termed 'mobile' and may exhibit malicious behaviour. We can trust few, if any, of the programs that we download. Mediating the behaviour of programs (whether we trust them or not) is a challenging task that involves making decisions based on criteria explicitly stated in a policy. The process of writing an effective policy is notoriously difficult, as we cannot predict the behaviour that a program will exhibit in the future; rather, we can only state what behaviours similar programs may have exhibited in the past. I present my ongoing work, which aims to move the problem of securely executing untrusted programs away from programmer-implemented security and into the domain of the 'end user'. That is, he or she who may not understand programming concepts, but can still outline the constraints they wish to place on the untrusted programs that they run. My work builds upon Polymer; a system that creates runtime monitors, which run in parallel with an untrusted program and examine actions invoked by the program's instruction stream. It adds a language for the high-level specification of policies, to enable them to be easily translated from plain English statements about program behaviour. Rules derived by the language can be temporally ordered and used to reason about the actions invoked by the program, as well as the flow of information through the program (the latter part is my most-recent work). Policies written in the language will be translated into Polymer policies by a compiler and used to create runtime monitors, prior to these being instrumented on the untrusted program. It is intended that the specification and compilation be driven by a GUI, to fully achieve 'end user' operability. Any action which is detected by a monitor to violate the policy may result in the application being halted, the individual action being suppressed or the action being replaced by a precomputed 'safe' action. This sees the creation of a program with 'trusted' behaviour, from an untrusted one. -------------------------------- Date and time: Monday 7th November 2005 at 13:00 Location: UG40, School of Computer Science Title: Introducing the Cake Talk Challenge: Predictive Text Speaker: David Brooks & Nick Hawes (http://www.cs.bham.ac.uk/~djb) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: In this talk we will discuss our idea of using the Cake Talk series to encourage basic collaborative research within the department using ongoing challenges. The idea is that a group of people propose a fairly general problem, and then researchers from different fields apply various approaches to solving it. Following this introduction, we will describe the first of these challenges: investigating ways of improving the T9 predictive text entry system. We will give an introduction to T9 itself, and an extremely high-level overview of some of the research problems it presents. After this we will present a Java toolkit for evaluating predictive text models, and some example systems that use it. We will give brief overview of how the toolkit can be used in a machine learning context. The toolkit will be freely available to enable others to take part in the challenge (and possibly a competition!). The talk will be informal and high-level (whilst hinting at the technical complexities of the problem), and discussion will be actively encouraged. -------------------------------- Date and time: Monday 14th November 2005 at 13:00 Location: UG40, School of Computer Science Title: Preliminaries to a specification language for Prelude Systems Speaker: Pascal Honore (http://www.cs.bham.ac.uk/~pxh) Institution: School of Computer Science, The University of Birmingham (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: In my last talk, last month, I did a demo of two languages which are the core of the Ink.Prelude Project. I have now written the entire documentation for this project which can be found here: http://www.cs.bham.ac.uk/~pxh/research/files/ink.prelude.pdf (Comments are welcome). Other material (such as the compilers) for this project can also be found on my School web-page. Though my PhD concerns essentially mobile systems, the Prelude framework can be used in various situations such as distributed systems, colonies of processes, agents interaction design etc... (I always forget to mention this is my abstracts: my work is about model checking, but for its use by other people, so feel free to get interested. You do not have to *be* a model checking person to know what automatic verification techniques, and in particular a high level frameworks such as Prelude, can do for you...) In this talk I will present a specification language for Prelude systems. This was the "missing" part. -------------------------------- Date and time: Monday 21st November 2005 at 13:00 Location: UG40, School of Computer Science Title: Why WikiWiki Web Works Well Speaker: Andy Pryke (http://www.cs.bham.ac.uk/~anp) Institution: School of Computer Science, The University of Birmingham (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: In this talk I'll give an overview of Wiki philosphy, outline potential problems and how they have been addresses, demonstrate the practical use of the Cercia Wiki, and make live updates to the worlds largest encyclopedia. I will also talk briefly about the options for setting up your own Wiki. WikiWiki systems [1] are just over 10 years old and have become increasingly widely used over the last few years. The idea behind Wiki is simple: every web page is editable, in the browser, straight away. It overcomes one of the key problems of traditional hypertext systems - that it's difficult to keep them current and correct. With a traditional one-webmaster system you would fire up an email client, cut and paste links to the pages needing changes, explain your corrections or additions and send off the email to the webmaster. With a Wiki system, if you see an error or a place where you can add information, you click on "edit", type some plain text, and click "save". The website is updated. This approach is particularly useful in traditionally information sharing cultures such as academia and computing. [1] http://en.wikipedia.org/wiki/Wiki -------------------------------- Date and time: Wednesday 14th December 2005 at 13:00 Location: UG40, School of Computer Science Title: Transcription, translation and in between: The road to complexity in artificial evolution Speaker: Philipp Rohlfshagen (http://www.cs.bham.ac.uk/~pzr) Institution: School of Computer Science, The University of Birmingham (http://www.cs.bham.ac.uk) Host: Nick Hawes Abstract: In this talk I will be introducing the Human Genome Project and discuss some interesting facts about the human genome. I will give a basic overview of genetics followed by a more detailed yet simple description of post-transcriptional processes. This is followed by a brief overview of my current work: An extended genetic algorithm that is based upon some biochemical processes regulating the expression of RNA. This overview implicitly poses the question whether (and to what degree) artificial evolution (genetic algorithms in particular) can benefit from the latest insights in genetics. I would like to make this question an open debate to discuss the benefits and drawbacks of such an approach and possibly identify some rough guidelines as to how to proceed. -------------------------------- Date and time: Monday 16th January 2006 at 13:00 Location: UG05, Learning Centre Title: Are DNA repair genes selfish? Speaker: Marie Bergem-Ohr Institution: Institute for Basic Medical Science, Faculty of Medicine, University of Oslo Host: Chrisantha Fernando Abstract: DNA repair genes encode proteins that promote stringent replication of genetic information. Different genes are related to repair of particular types of DNA modification in specific sequences. In line with the selfish gene hypothesis we therefore predict that DNA repair genes have evolved to protect their own sequence more than other sequences. Viewed in a different perspective, we expect that particular nucleotide sequences have infiltrated the DNA repair genes that protect them from mutations. Our recent study of DNA mismatch repair (MMR) genes has confirmed this hypothesis (Breivik and Falster, submitted). Evidence from cancer research demonstrates that MMR genes introduce an insertion bias in mononucleotide repeats, thereby promoting their formation and persistence though the course of evolution. Accordingly, we predicted that mononucleotide repeats should be particularly abundant in and around MMR genes. We therefore performed a complete mapping of mononucleotide repeats in the genome of several different organisms, and our hypothesis was strongly confirmed. We now aim to explore and test this model in a broader perspective, involving other DNA repair genes and their target sequences. We are particularly interested in genes related to the stability of trinucleotide repeats, CNG repeats. These sequences are intrinsically unstable due to the formation of secondary structures, and play a key role in the pathogeneses of cancer and several neurological conditions, including Huntington’s disease. As for MMR, our preliminary results suggest that CNG repeats are particularly abundant in and around specific repair genes, and we are setting up a database that organize this relationship between DNA repair genes, their phenotypic function and their distribution of particular sequence motifs. We believe this research and the underlying model may provide an evolutionary explanation for the origin of unstable sequences in key genes throughout the genome. The concept may have implication for the understanding of several diseases, and suggest a previously unrecognized evolutionary principle by which a genes DNA modifying phenotype may specifically shape its own genotype. -------------------------------- Date and time: Tuesday 14th February 2006 at 13:00 Location: UG40, School of Computer Science Title: Intelligent Assistance for Model Building in the Social Sciences using Data Driven Simulation Speaker: Catriona Kennedy (http://www.cs.bham.ac.uk/~cmk) Institution: School of Computer Science (http://www.cs.bham.ac.uk) Abstract: "Data driven" simulation is a process by which simulations are continually adjusted to reality. The simulaton's predictions can also play a role in selecting the aspects of reality to be observed. We consider two different classes of intelligent agent that can control a data driven simulation: (a) an autonomous agent using internal simulation to test and refine a model of its environment and (b) an assistant agent managing a data-driven simulation to help humans understand a complex system (assisted model-building). In the first case the agent is situated in its environment and can use its own sensors to explore the data sources. In the second case, the agent has much less independent access to data and may have limited capability to refine the model on which the simulation is based. This is particularly true if the data contains subjective statements about the human view of the world, such as in the social sciences. For complex systems involving human actors, we propose an architecture in which assistant agents cooperate with autonomous agents to build a more complete and reliable picture of the observed system. -------------------------------- Date and time: Tuesday 28th March 2006 at 13:00 Location: UG40, School of Computer Science Title: Using high-level security policies to transform unsafe programs into safe programs Speaker: Andy Brown (http://www.cs.bham.ac.uk/~ajb) Institution: School of Computer Science, The University of Birmingham (http://www.cs.bham.ac.uk) Abstract: Users often execute programs that they do not trust. Such programs offer useful functions but can also contain malicious behaviour. The user must decide to trust the program and risk it causing damage to their system, or not to trust it and to use some additional technology to attest it, or to harness its execution. Digitally signed code, proof-carrying code, and firewalls can prevent program execution from causing damage, but have deficiencies themselves. Digitally signed code, for example, may enhance user confidence that an executable has not been altered, but still requires the user to trust the code signee. Program transformation can assist the problem of running untrusted code; it allows the user to transform it into safe code that complies with a security policy. Our work builds on Polymer: a system for policy enforcement through the transformation of Java programs. A policy specifies action sequences that should not be invoked by a program, before detailing the actions (and their parameters) that should be inserted into the program when a policy violation is detected. If program transformation is to be used to prevent malware, the techniques it offers must be operable by the end-user (constructing an effective policy that is guaranteed to meet a set of security-critical requirements is difficult for any non-programmer). Our work aims to develop an end-user operable anti-malware system with a high-level policy specification language (which may be driven by a user interface) and a compiler into Polymer. In this talk we consider examples of malicious applications and demonstrate their transformation into variants that comply with a security policy. Current logics that policy specification uses are too weak to represent some properties of the programs that we may wish to monitor. We discuss the theoretical foundations that our work builds upon and detail our investigations into suitably expressive logics for policy specification. We conclude by stating our ideas about the format of a high-level language to achieve end-user policy specification at the highest feasible level. ------------ Further details:-- This talk is to be given at the 6th Analysis, Slicing and Transformation Network workshop (Security studies) on 3rd - 4th April 2006 (See: http://www.dcs.kcl.ac.uk/staff/zheng/astrenet/html/astrenet6.html) It is designed to last approximately 30 minutes. After the talk I would like to encourage a discussion on one or more of the following:-- * What do talk attendees currently see as the highest feasible level at which a user can specify a security policy? * From a purely conceptual viewpoint, how do the talk attendees think malware will be prevented in the future? * Which is preferable? Program execution governed by policies which are written by: an end-user, a trusted third party, or an intelligent agent? -------------------------------- Date and time: Tuesday 27th June 2006 at 13:00 Location: UG40, School of Computer Science Title: Alternative Splicing in Evolutionary Computation: Adaptation in Dynamic Environments Speaker: Philipp Rohlfshagen (http://www.cs.bham.ac.uk/~pzr/) Institution: School of Computer Science, The University of Birmingham (http://www.cs.bham.ac.uk) Abstract: In this talk, we present a new technique to efficiently solve a dynamic version of the 0/1 knapsack problem. The talk will cover some basic genetics including a more detailed description of alternative splicing, a biochemical process that modifies RNA to produce multiple proteins from a single template of DNA. Alternative Splicing is strongly associated with multi-cellularity and is thought to be essential to the evolution of complex organisms. We use this process as inspiration to design a genetic algorithm that tracks changes in a dynamic environment. Tracking is done using a combination of explicit and implicit memory facilitated by a modular encoding that is controlled by an abstract implementation of alternative splicing. The talk also covers the importance of a gene's modular composition and the regulation of such modules in the information processing architecture of the cell. -------------------------------- Date and time: Tuesday 18th July 2006 at 13:00 Location: UG40, School of Computer Science Title: Functional Programming on the Web Speaker: Noel Welsh (http://www.cs.bham.ac.uk/~nhw) Institution: School of Computer Science, The University of Birmingham (http://www.cs.bham.ac.uk) Abstract: Continuations, functional reactive programming, and bidirectional programming. A random walk down http://lambda-the-ultimate.org/ or the next Big Thing in web development? In the long and glorious tradition of Cake Talks I will present some half-baked ideas that argue for the later interpretation. Turn up and decide for yourself. Background: What are continuations, and why might they be useful for web programming? http://www.double.co.nz/pdf/continuations.pdf http://www.ccs.neu.edu/scheme/pubs/esop2003-gfkf.pdf http://www.seaside.st What is functional reactive programming?: http://www.haskell.org/frp/ http://www.cs.brown.edu/~greg/ What is bidirectional programming?: http://www.cis.upenn.edu/~bcpierce/papers/index.shtml -------------------------------- Date and time: Tuesday 22nd August 2006 at 13:00 Location: UG40, School of Computer Science Title: An Exonic Genetic Algorithm with RNA Editing Inspired Repair Function for the Multiple Knapsack Problem Speaker: Philipp Rohlfshagen Abstract: In this talk, we present a new approach to efficiently solving instances of the multiple knapsack problem. Our implementation uses an adaptive repair technique based upon a biochemical process found in cells: invalid encodings are repaired using a process analogous to RNA editing. RNA editing is a post-transcriptional process that selectively targets individual nucleotides to restore functionality of an otherwise dysfunctional protein. In a similar fashion, to restore the encoding's validity, our repair technique inverts individual bits in an order that evolves during the evolutionary run. This adaptive approach performs superior to other approaches in the literature, including those making explicit use of instance specific information. -------------------------------- Date and time: Tuesday 22nd May 2007 at 13:00 Location: UG40, School of Computer Science Title: Genetic Algorithms Tailored to Specific NP-Complete Optimisation Problems Speaker: Philipp Rohlfshagen (http://www.cs.bham.ac.uk/~pzr/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: In this cake talk, I will give two 20-minute presentations about 2 distinct yet conceptually related approaches in the field of evolutionary computation. Both approaches use genetic algorithms tailored to a specific NP-complete optimisation problem. The inspiration for the design of these algorithm's is drawn from the field of molecular genetics and it is shown how molecular processes may effectively be utilised in the field of combinatorial optimisation. The talks, which will later be presented at GECCO'07, are as follows: Title 1: A Genetic Algorithm with Exon Shuffling Crossover for Hard Bin Packing Problems A novel evolutionary approach for the bin packing problem (BPP) is presented. A simple steady-state genetic algorithm is developed that produces results comparable to other approaches in the literature, without the need for any additional heuristics. The algorithm's design makes maximum use of the principle of natural selection to evolve valid solutions without the explicit need to verify constraint violations. Our algorithm is based upon a biologically inspired group encoding which allows for a modularisation of the search space in which individual sub-solutions may be assigned independent cost values. These values are subsequently utilised in a crossover event modelled on the theory of exon shuffling to produce a single offspring that inherits the most promising segments from its parents. The algorithm is tested on a set of hard benchmark problems and the results indicate that the method has a very high degree of accuracy and reliability compared to other approaches in the literature. Title 2: ExGA II: An Improved Exonic Genetic Algorithm for the Multiple Knapsack Problem ExGA I, a previously presented genetic algorithm, successfully solved numerous instances of the multiple knapsack problem (MKS) by employing an adaptive repair function that made use of the algorithm's modular encoding. Here we present ExGA II, an extension of ExGA I that implements additional features which allow the algorithm to perform more reliably across a larger set of benchmark problems. In addition to some basic modifications of the algorithm's framework, more specific extensions include the use of a biased mutation operator and adaptive control sequences which are used to guide the repair procedure. The success rate of ExGA II is superior to its predecessor, and other algorithms in the literature, without an overall increase in the number of function evaluations required to reach the global optimum. In fact, the new algorithm exhibits a significant reduction in the number of function evaluations required for the largest problems investigated. We also address the computational cost of using a repair function and show that the algorithm remains highly competitive when this cost is accounted for. -------------------------------- Date and time: Tuesday 5th June 2007 at 13:00 Location: Room 124, School of Computer Science Title: Principles of Model Driven Software Engineering (MDSE) and tools for visualisation Speaker: Seyyed Shah (http://www.cs.bham.ac.uk/about/people/person.php?group=2&name=szs) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: (NOTE: There will be a brief, 5 minute, warm-up talk by Damien on his use of Models in Computer Vision) "Model Driven Software Engineering (MDSE) is an emerging methodology for integrated software development. For the software engineer its potential advantages are very attractive: rapidly developed high quality systems, greater scope for code reuse, lower complexity through abstraction and providing an overall “productive environment”. This informal talk will try to introduce the principles and my area of research in MDSE. The talk aims to define the role of important concepts such as “Modelling”, “Meta-Modelling”, “Model Transformation” to non practitioners. Some specific work on developer support in MDSE will also be introduced. A simple visualisation framework was developed as part of this research. It can draw UML-inspired Class, Object and Sequence Diagrams from transformation code. The general goal of the framework is to facilitate developer understanding of model transformations. Sample diagrams and their use will be presented, time permitting. (This talk is in practice for presenting a research poster to non-computer scientists, for the universities' UK-Grad poster competition)" -------------------------------- Date and time: Tuesday 12th June 2007 at 15:00 Location: UG40, School of Computer Science Title: Smart Cards - Security and Applications Speaker: Hasan Qunoo (http://www.cs.bham.ac.uk/~hxq/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: Smart cards have been used in the past decades in many application areas. Mobile telephones, credit cards, electronic purses, public card telephones, medical insurance cards and personal identity cards are typical smart card applications, and there are many more. In order to produce smart card applications, the creation of national and international standards to specify smart cards interface and specification were necessary. Smart cards are used as data storage media, authorization media and encryption modules. Practical methods and procedures for cryptography, key exchange and trust protocols have been developed to suit smart card specifications. Smart cards are used to establish trust in sensitive applications. This trust is ensured by the appropriate design of four components of smart card security: the card body, chip hardware, operating system and application. In this informal talk, I will establish an overview of the smart card specification and standards. I will discuss the methods and procedures used to provide security and describe some of the techniques of smart card security engineering. I also will use electronic payments in smart cards as a case study. -------------------------------- Date and time: Friday 14th September 2007 at 12:00 Location: UG40, School of Computer Science Title: Challenges for Planning for Autonomous Underwater Vehicles Speaker: Zeyn Saigol (http://www.cs.bham.ac.uk/~zas/) Abstract: This 15-minute talk is in preparation for a workshop on real-world planning at ICAPS '07 (http://www.mbari.org/autonomy/ICAPS07-workshop/). The paper it is based on is available at http://www.cs.bham.ac.uk/~zas/ICAPSworkshop07.pdf Hydrothermal vents are spectacular volcano-like eruptions found at the bottom of oceans. Driven by plate tectonics, they are of great importance to marine scientists, as they provide clues to the inner working of the Earth, and support a variety of exotic lifeforms. Traditionally, manned or remotely-operated submersibles have been used to hunt out hydrothermal vents on the sea floor. However, there are several benefits to having an unmanned, completely autonomous submersible search for hydrothermal vents, such as lower engineering and operating costs, and longer mission durations. In this talk, I introduce the problem of finding hydrothermal vents by detecting chemical tracers that are emitted from the vent sites. From a planning perspective, the problem has several features that make it challenging: the location of vents is not directly observable, as tracers only provide a partial clue to the whereabouts of their source; the state and action spaces are continuous; there are limited resources for executing the plan; and the action outcomes are uncertain. While planners exist that can deal with each of these independently, none are able to handle all of them at once. I will also (very briefly) sketch possible approaches for solving the problem. A key step is to separate the mapping and planning problems and tackle each separately. -------------------------------- Date and time: Tuesday 18th September 2007 at 13:00 Location: UG40, School of Computer Science Title: Evolving Neuromodulatory Topologies for Reinforcement Learning-like Problems Speaker: Andrea Soltoggio (http://www.cs.bham.ac.uk/~aqs/) Abstract: Environments with varying reward contingencies constitute a challenge to many living creatures. In such conditions, animals capable of adaptation and learning derive an advantage. Recent studies suggest that neuromodulatory dynamics are a key factor in regulating learning and adaptivity when reward conditions are subject to variability. In biological neural networks, specific circuits generate modulatory signals, particularly in situations that involve learning cues such as a reward or novel stimuli. Modulatory signals are then broadcast and applied onto target synapses to activate or regulate synaptic plasticity. Artificial neural models that include modulatory dynamics could prove their potential in uncertain environments when online learning is required. However, a topology that synthesises and delivers modulatory signals to target synapses must be devised. So far, only handcrafted architectures of such kind have been attempted. Here we show that modulatory topologies can be designed autonomously by artificial evolution and achieve superior learning capabilities than traditional fixed-weight or Hebbian networks. In our experiments, we show that simulated bees autonomously evolved a modulatory network to maximise the reward in a reinforcement learning-like environment. -------------------------------- Date and time: Tuesday 20th November 2007 at 13:00 Location: Coffee Room Title: Informal Discussion Session Speaker: Everyone Abstract: For the research students to get together and discuss - ideas, research, etc. It's about collegiality, dude. -------------------------------- Date and time: Tuesday 27th November 2007 at 13:00 Location: Coffee Room Title: Informal Discussion Session Speaker: Everyone Abstract: For the research students to get together and discuss - ideas, research, etc. It's about collegiality, dude. -------------------------------- Date and time: Tuesday 4th December 2007 at 15:00 Location: Coffee Room, Level 1, School of Computer Science Title: Research Summaries Speaker: Peter Lewis & Zeyn Saigol Institution: School of Computer Science Abstract: Zeyn & Peter will tell us what they're working on at the moment. Zeyn is working on underwater autonomous robots and Peter has a poster. -------------------------------- Date and time: Tuesday 11th December 2007 at 15:00 Location: Coffee Room Title: Informal Discussion Session Speaker: Everyone Abstract: For the research students to get together and discuss - ideas, research, etc. It's about collegiality, dude. -------------------------------- Date and time: Tuesday 18th December 2007 at 15:00 Location: Room 245, School of Computer Science Title: Dynamic Optimization & Computer Vision Speaker: Philipp Rohlfshagen & Damien Jade Duff (http://www.cs.bham.ac.uk/~pzr/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: (NOTE: because room 124 is booked, we're using room 245, and also note that the seminar time is 3pm) Dynamic optimization is the term broadly applied to approaches to optimization problems where the problem to be optimized is changing in some structured way with time. Philipp is starting out on some original research in the area of dynamic optimization, particularly from a natural computation background. Computer Vision pose-estimation problems are typically formulated as optimization (or "adjustment") problems, usually to minimize some error metric in the estimate of object pose. Damien has some limited knowledge of this area. Pose-estimation over time (or "tracking") is thus a dynamic optimization problem. Together they'll (attempt to) tackle (some of) the (rather) slippery ground between this set of techniques (dynamic optimization) and this application area (visual tracking). -------------------------------- Date and time: Tuesday 8th January 2008 at 15:00 Location: Room 245, School of Computer Science Title: Why has Artificial Intelligence not succeeded? Speaker: Zeyn Saigol (http://www.cs.bham.ac.uk/about/people/showperson.php?person_id=3716) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: The aim of this cake talk is mostly to stimulate discussion on the topic of AI, and what its future directions should be. I don't have any deeply-held views on why the strong AI goal has not yet been achieved, and I'm not going to cover any of the philosophical arguments on whether or not this goal is even possible. Instead, I intend to sketch a quick outline of the progress of AI over the last 50 years, and some possible reasons the AI problem has turned out to be harder than anyone expected. -------------------------------- Date and time: Tuesday 15th January 2008 at 15:00 Location: Room 245, School of Computer Science Title: The Perils of Multi-core programming - Why things don't just get faster Speaker: Michael Kneebone (http://www.cs.bham.ac.uk/~mlk/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: Multi-core machines - computers with more than 1 processing core - are now mainstream in desktop and laptop hardware. Contrary to previous CPU improvements, the programmers "free-lunch" for speed-improvements is evaporating as code must be explicitly parallel to utilise multiple cores. In this talk I explain the difficulties in doing so, both practically and theoretically, and explain what can happen at the sub-CPU level to aid the programmer. I'm also going to talk about the internal structure of a modern CPU and where a good idea can sometimes work against you. -------------------------------- Date and time: Tuesday 22nd January 2008 at 15:00 Location: Room 124, School of Computer Science Title: Verifying Access Control Policies through Model Checking Speaker: Hasan Qunoo (http://www.cs.bham.ac.uk/~hxq/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: The Internet is about sharing. Today's systems enable you to socialise, share your photos, shop for your favourite music album, and watch a film online. When users subscribe to those services, they agree to participate, often implicitly, in the systems access policy. As the services provided become more sophisticated, the policy becomes more dynamic and dependant on other participants actions. Most of the time, the user is only an actor in long chain of actors involved in a complex highly decentralised procedure. The user has to trust, not only the interface service provider, but all the participants, without being able to evaluate and reason about the system policy. This has encouraged a number of research groups around the world to develop methods and techniques for expressing and enforcing access control policy and thus how we might model policies and reason about them. In this talk I explain how can we use Model Checking tecniques to model and reason about access control systems and why it is difficult to model large access control policies. I'm also going to discuss the ways in which I am planing to improve our ability to express concepts, like: integrity constraints and delegation of authorities. -------------------------------- Date and time: Tuesday 29th January 2008 at 15:00 Location: Room 245, School of Computer Science Title: Memory in Dynamic Optimisation Speaker: Philipp Rohlfshagen (http://www.cs.bham.ac.uk/~pzr/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: The use of memory is a somewhat popular technique used to enhance the performance of evolutionary algorithms (EAs) in dynamic optimisation problems (DOPs). The implementation of memory is either implicit (part of the encoding) or explicit (external register). Explicit memory can be hugely beneficial in cyclic domains yet it seems a significant proportion of publications also uses such memory in non-cyclical domains where potential benefits are not clear at all. This talk will first present a brief overview of the field (implicit and explicit memory) followed by a discussion of the problems faced by explicit memory schemes. Finally, it will be argued that memory is indeed vital for EAs to do well in DOPs, although in a very different manner than previously assumed. -------------------------------- Date and time: Tuesday 5th February 2008 at 15:00 Location: Room 124, School of Computer Science Title: Informal Discussion Session Speaker: Research Students Abstract: For the research students to get together and discuss - ideas, research, etc. It's about collegiality, dude. -------------------------------- Date and time: Tuesday 19th February 2008 at 13:00 Location: UG40, School of Computer Science Title: Structured Peer-to-Peer Networks Speaker: Richard Price, Tien Tuan Anh Dinh & Robert Minson (http://www.cs.bham.ac.uk/~ttd/current_research.php) Institution: School of Computer Science (http://www.cs.bham.ac.uk/research/) Abstract: Peer-to-Peer (P2P) overlays are a class of distributed data structure infamous for their application within legally questionable file-sharing networks. Recent research has focused on introducing structure to such systems to transform them in to flexible, efficient and scalable distributed search platforms. Our talk will introduce these 'Structured Overlay Networks' and discuss some of the fundamental and applied research issues surrounding them. This talk is aimed at a general audience with no prior knowledge of distributed data structures. It will appeal to anyone with an interest in Distributed Systems, Self-Organising Systems, Social Networks, Security and pretty much anything else, particularly cakes. -------------------------------- Date and time: Tuesday 26th February 2008 at 15:00 Location: Room 124, School of Computer Science Title: Probability density modelling of Galaxies from data with complex structures Speaker: Xiaoxia Wang Institution: School of Computer Science Abstract: The distribution of galaxies during their evolution is represented by a particle model in astronomy, which could raise some problems when carrying out the comparison between two simulations or between a simulation and real observations. To overcome these problems, we consider using probability density modelling to represent the distribution of galaxies. Galaxies might generate very structured distributions during their evolution. Capturing these structures could help us to build a more accurate model and also reveal information about the galaxies evolutionary progress. We propose a framework to discover these structures and to build a hierarchical probability density model to align them. Because there are more than one structure (or "manifold") embedded in the data, the discovery problem offers a great challenge to the existing manifold learning approaches, which can only learn one manifold. In our work, we developed an algorithm to learn all these structures by estimating gradients on these structures or manifolds. -------------------------------- Date and time: Tuesday 4th March 2008 at 15:00 Location: Room 124, School of Computer Science Title: Informal Research Discussion Session Speaker: Research students (http://www.cs.bham.ac.uk/) Abstract: For the research students to get together and discuss - ideas, research, etc. It's about collegiality, dude. -------------------------------- Date and time: Tuesday 18th March 2008 at 15:00 Location: Room 124, School of Computer Science Title: An Introduction to Reflectance-Based Diagnosis Speaker: Amaria Zidouk Institution: School of Computer Science, University of Birmingham Abstract: As a result of recent developments in modern imaging devices, medical images bestow uniquely detailed views of internal anatomy. Such views can yield large amounts of information that require the use of computers to analyse it. To this aim, efforts were exerted to develop computer-based models of the optical properties of tissues as part of reflectance-based optical diagnosis methods. Such techniques can be implemented without the need for invasive biopsies and have the potential to improve the early detection of changes in tissues. I will demonstrate how modelling the optical properties of human skin was successfully used in the diagnosis of malignant melanoma and how optical diagnosis techniques can contribute to the early detection of oesophageal cancer. -------------------------------- Date and time: Tuesday 22nd July 2008 at 13:00 Location: UG40, School of Computer Science Title: OBIMOD Open-ended Bio-Inspired Modelling and Design Workshop Speaker: Abstract: Strongly bio-inspired models and design procedures draw inspiration from biological processes, physiological mechanisms, neural dynamics and other complex phenomena observed in nature. Unfortunately, bio-inspired notions are often quickly applied to and validated on optimisation problems, resulting in a short-term struggle for performance that forsakes the original inspiring principles. Releasing the focus from the performance, OBIMOD researchers enjoy more freedom to explore original ideas, novel approaches and devote more attention to the inspiring sources. Open-ended bio-inspired modelling and design investigate ideas like gene regulatory networks and artificial genomes, developmental processes, scalable and novel neural systems, generative, robust, fault-tolerant and immune systems, artiticial behaviour and Alife, learning, memory and other. This workshop was organised to bring together OBIMOD researchers at the School of Computer Science. Please note that the following programme is not final. Programme: 13:00 - 13:05 Andrea Soltoggio - Introduction 13:05 - 13:30 Ben Jones - Evolving Functional Symmetry in a Three Dimensional Model of an Elongated Organism Abstract - In any organism having a nervous system, a rich coupling exists between the nervous system and the organism's body-plan (think of a jelly fish compared to a flatworm). During evolution, this coupling evolved in a complex fashion resulting in a divergence of body-plan symmetries. In a given fish-swimming type niche, bilateral symmetry is advantageous to the fish, but the coupling underpinning this advantage is less clear. To make it clearer, we constructed a model of an eel-like organism and evolved both the motor configuration (which we considered to be part of the body-plan) in concert with the nervous system (nb, a ctrnn controller). Although the modelled `organism' is no trout, simulated evolution typically finds motor configurations that are bilaterally symmetric. This signifies an importance in bilateral functionality allowing us to clarify the overall advantage of bilateral symmetry for long elongated organisms. 13:30 - 13:50 Victor Landassuri - A new Approach for Incremental Modular Neural Networks in Time Series Forecasting Abstract - Modular Neural Networks have been used to solve complex problems in a reduced amount of time, to obtain better performance on a range of tasks, and to provide a better understanding of the human brain. In this talk I will present the first stage of my research into Modular Neural Networks where the evolution of them allows an incremental architecture for solving more than one problem. The basis for developing that approach is described with some related issues in Time Series Forecasting using an Evolutionary Algorithm called EPNet. 13:50 - 14:10 Ed Robinson - tba Abstract - tba 14:10 - 14:35 Thomas Miconi - Fitness Transmission: A Genealogic Signature of Adaptive Evolution Abstract - We introduce Fitness Transmission as a simple statistical signature of adaptive evolution within a system. Fitness transmission is the correlation between the fitness of parents and children, where fitness is evaluated after the number of grandchildren, suitably normalised. This measure is a direct calculation based on a genealogical record, rather than on genetic or phenotypic observation. We point out that the Bedau-Packard statistics of evolutionary activity cannot be used as a reliable system-wide signature of adaptive evolution, because they can produce positive signals when applied to certain ``random'', non-evolutionary systems. We apply fitness transmission to simple evolutionary algorithms (as well as neutral equivalents) and demonstrate its capacity to accurately detect the presence or absence of Darwinian evolution. 14:35 - 14:50 Break 14:50 - 15:05 Andrea Soltoggio - An Introduction to Analog Genetic Encoding (AGE) Abstract - In biology, phenotypical features derive from a complex interactions of genes. Analog Genetic Encoding (AGE) is an encoding method developed at EPFL based on the idea of an artificial genome. AGE has been used to encode gene regulatory networks (GNRs), electronic circuits and other graphs like neural networks. The main features of AGE and how-tos will be covered in a short introduction. 15:05 - 15:25 Chris Bowers - An Introduction to Computational Embryogeny Abstract - Developmental processes clearly have a huge influence on biological systems yet the complex dynamics they introduce are poorly understood. In this talk I will discuss the relationship between developmental process and evolution utilising a simple computational model of embryogeny. 15:25 - 15: 40 Victor Landassuri - Special talk - Intelligent Recycling Station, a project for Imagine cup 2008 Abstract - In this talk I am going to describe an interactive device that helps with the task of recycling materials, reducing objects and trash in a home scenario. The project was developed with other two partners to participate in a Microsoft's competition called "Imagine cup 2008" in the embedded development category. Even though we started the project as hobby, we could pass to the final, so I will state too, some remarkable aspects for anyone interested in participate in this kind of competitions. 15:40 - 16:00 Andrea Soltoggio - Advantages of Neuromodulated Plasticity Abstract - Neuromodulation is considered a key factor for learning and memory in biology. We test this hypothesis in artificial neural networks by introducing a new type of neuron: modulatory neurons. Simulated evolution designs neural control networks for learning problems. The results show that modulatory neurons help achieving better learning. We conclude that modulatory neurons evolve autonomously in the proposed learning tasks, allowing for increased learning and memory capabilities. 16:00 : Late breaking talks and discussions -------------------------------- Date and time: Thursday 22nd January 2009 at 14:00 Location: Room 124, School of Computer Science Title: AI Planning for Robotic Submarine Exploration Speaker: Zeyn Saigol (http://www.cs.bham.ac.uk/~zas/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: In this talk, I will present a high-level discussion of my PhD research, including some recent results from experimental simulations, plus an overview of a "working holiday" I spent in California actually working with underwater robots. Marine exploration is interesting for AI because it requires us to deal with a very large state space and highly uncertain environment. The aim of my PhD is to develop techniques for Autonomous Underwater Vehicles (AUVs) to find hydrothermal vents more efficiently, so I will start by describing these vents, and some of the challenges of the environment. I will explain the methods I've used to provide an initial solution the problem, which rely on looking a short distance into the future and estimating the value of actions beyond that. I should be able to show some results and videos of the algorithms in action. I'll also discuss my work at MBARI in California, and may show some pictures from my trip including ones of their robots. My project there focused on automated testing methods for the planning software they use on their AUVs. -------------------------------- Date and time: Thursday 29th January 2009 at 15:00 Location: Room 124, School of Computer Science Title: Bi-objective Optimization for the Vehicle Routing Problem with Time Windows Speaker: Abel Garcia Najera (http://www.cs.bham.ac.uk/~agn/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: The Vehicle Routing Problem with Time Windows is a complex combinatorial optimization problem which can be seen as a fusion of two well known sub-problems: the Travelling Salesman Problem and the Bin Packing Problem. Its main objective is to find the lowest-cost set of routes to deliver demand, using identical vehicles with limited capacity, to customers with fixed service time windows. In this talk I will present an overview of my research, including results from my paper that has been accepted in EMO'09 (http://www.emo09.org/), in which we propose a method to measure route similarity and incorporate it into an evolutionary algorithm to solve the bi-objective VRPTW. -------------------------------- Date and time: Thursday 12th March 2009 at 14:00 Location: Room 124, School of Computer Science Title: Towards a New Synchronous Programming Language Speaker: Mohamed Nabih Menaa (http://www.cs.bham.ac.uk/~mnm) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: In this talk, I will present an outline of my current research on synchrony in programming languages. Synchronous programming languages are a family of languages that share a common mathematical foundation. Essentially, programs in such languages proceed in a sequence of atomic steps where computational micosteps can occur simultaneously. We project to design such a language along with a formal (game) semantics. Our approach relies on a canonical model of compositional systems (a monoidal closed category) to interpret programming language constants. I will outline the goals and motivation of my research, introduce the semantics framework to be used, and then give a few examples of language constants interpretations. This talk is intended for a general audience, and therefore, no previous familiarity with the topic or its background will be assumed. -------------------------------- Date and time: Thursday 18th June 2009 at 14:00 Location: Room 124, School of Computer Science Title: A Linear Grammar Approach to Mathematical Formula Recognition from PDF Speaker: Josef Baker (http://www.cs.bham.ac.uk/~jbb/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: Many approaches have been proposed over the years for the recognition of mathematical formulae from scanned documents. More recently a need has arisen to recognise formulae from PDF documents. Here we can avoid ambiguities introduced by traditional OCR approaches and instead extract perfect knowledge of the characters used in formulae directly from the document. This can be exploited by formula recognition techniques to achieve correct results and high performance. In this talk I will present an overview of my research, including results from our paper that has been accepted in MKM'09 (http://www.orcca.on.ca/conferences/cicm09/mkm09/) in which we revisit an old grammatical approach to formula recognition, that of Anderson from 1968, and assess its applicability with respect to data extracted from PDF documents. -------------------------------- Date and time: Thursday 25th June 2009 at 14:00 Location: Room 124, School of Computer Science Title: Neural Network Ensembles for Time Series Forecasting Speaker: Victor Landassuri (http://www.cs.bham.ac.uk/~vlm/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: The evolution of Artificial Neural Networks is an area that has been widely studied in the past years giving several different algorithms to evolve them. On example of that is the EPNet algorithm designed over the Fogel’s Evolutionary Programming approach to evolve small network's architectures. However, the EPNet algorithm had been tested in a reduced number of time series (TS), being the classification task the main approach to test the performance of the algorithm. It this work I will present the improvements already done for the EPNet Algorithm as well the study carried out with two ensemble approach – Average and Rank Base Linear combination ensembles. To measure the performance of the algorithm (and to rectify the deficiency presented above) it was used several TS of different behaviours and dynamics in the time series forecasting task. The results of this presentation will be given in the GECCO'09 conference, so any comment or suggestion to improve it will be gratefully appreciated. -------------------------------- Date and time: Monday 26th October 2009 at 14:00 Location: Room 124, School of Computer Science Title: Infinite data structures, computability and decidability Speaker: Olaf Klinke (http://www.cs.bham.ac.uk/~okk/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: I will try to convince the audience that topology, which is a field of mathematics arising from geometry, is useful in computer science when one does computing with infinite data structures such as streams of numbers, infinite trees and alike. In particular we discuss different methods of representing real numbers and relate them to geometric ideas. We show how geometric ideas translate quite naturally to issues about deciding properties of numbers such as 'is the number positive?' or 'is the number equal to 0?' or 'can this function be implemented in some programming language?' -------------------------------- Date and time: Thursday 5th November 2009 at 14:00 Location: Room 124, School of Computer Science Title: Use of market-based controls to simulate self-organization amongst web-applications, in the cloud Speaker: Vivek Nallur (http://www.cs.bham.ac.uk/~vxn851/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: There is an increasing interest in self-organization of software, due to properties like robustness and adaptibility of self-organizing biological systems. These are properties that we would like to imbue in our software, but have no reliable mechanism to do it with. Among software systems, there are broadly two kinds of change that we would like: 1) functional - the software changes functionality on its own 2) quality attributes - the software changes the quality of the functionality provided, on its own This talk will focus on self-organization needed to effect quality attribute changes, and our approach for self-organization. We investigate the use of market-based controls (demand, supply, price etc.) to effect self-organization. -------------------------------- Date and time: Thursday 19th November 2009 at 14:00 Location: Room 124, School of Computer Science Title: Game Theory: From Economics to Biology to Computer Science... And Back Again. Speaker: Peter Lewis (http://www.cs.bham.ac.uk/~prl/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: In this talk I shall introduce game theory, the theory of strategic interactions between rational individuals, and how the outcomes of these interactions relate to both the individuals’ preferences and the structure of the game. I shall outline a potted history of game theory, including its roots in attempts to reason about economic situations, its formalisation in the early part of the twentieth century, the development of the concept of Nash equilibria in the 1950s, and the theory’s extension as a tool for the analysis of evolutionary and other population based systems, mainly in the 1970s and 1980s. I will describe how computer science has enabled game theory and game theoretic systems to be investigated and applied in new ways, and how in turn this is solving problems in the domains of politics, economics, biology and computer science. -------------------------------- Date and time: Thursday 26th November 2009 at 14:00 Location: Room 124, School of Computer Science Title: Game Theory: From Economics to Biology to Computer Science... And Back Again Speaker: Peter Lewis (http://www.cs.bham.ac.uk/~prl/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: In this talk I shall introduce game theory, the theory of strategic interactions between rational individuals, and how the outcomes of these interactions relate to both the individuals’ preferences and the structure of the game. I shall outline a potted history of game theory, including its roots in attempts to reason about economic situations, its formalisation in the early part of the twentieth century, the development of the concept of Nash equilibria in the 1950s, and the theory’s extension as a tool for the analysis of evolutionary and other population based systems, mainly in the 1970s and 1980s. I will describe how computer science has enabled game theory and game theoretic systems to be investigated and applied in new ways, and how in turn this is solving problems in the domains of politics, economics, biology and computer science. -------------------------------- Date and time: Tuesday 19th October 2010 at 15:00 Location: Room 124, School of Computer Science Title: A Parallel Matching Algorithm for Continuous Interest Management Speaker: Elvis Liu Institution: School of Computer Science Abstract: In recent years, the scale of distributed virtual environments (DVEs) as grown rapidly in terms of number of participants and virtual entities. Many DVEs employ interest management schemes to reduce bandwidth consumption and thus enhance the scalability of the system. Most of the existing interest management approaches, however, have a fundamental disadvantage - they perform interest matching at discrete time intervals. As a result, they would fail to report events between consecutive time-steps of simulation which leads to incorrect simulations. This talk presents an interest matching algorithm which aims to capture missing events between discrete time-steps. This algorithm facilitates parallelism by distributing the workload of matching across multiple processors. Since it is increasingly common to deploy commercial DVE applications on shared-memory multiprocessor machines, using the parallel algorithm for these applications would be more suitable than the existing serial algorithms. -------------------------------- Date and time: Wednesday 2nd March 2011 at 12:00 Location: Room 124, School of Computer Science Title: Random triangles and the curse of dimensionality Speaker: Bob Durrant (http://www.cs.bham.ac.uk/~durranrj/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: Prerequisites: Passing familiarity with basic probability and the ideas of an integral, a limit, and 3D geometry. Outline: Given three vectors chosen randomly from a standard normal distribution in R^2, what is the probability that the triangle they form is acute or obtuse? It turns out that the probability of an obtuse triangle is exactly 3/4, and this can be shown using either a slightly scary integral or via a simple "proof by pictures" argument. Now what about if we choose three vectors randomly from a standard normal distribution in some outrageously high dimensional space R^d? Say R^1000 or R^10000? What is the probability that our random triangle is still obtuse? We can no longer draw the picture, so we revert to our scary integral and find that the probability of an obtuse random triangle drops sharply as d increases. In fact in the limit when d -> infinity the probability is zero. Why should this be? It turns out to be the result of a very general phenomenon in high dimensional space that the random vectors are very likely to have nearly the same length and be nearly orthogonal to each other. This means that in the limit our random triangle is equilateral with probability 1. Finally, these last facts are aspects of the "curse of dimensionality" and have practical implications for a range of popular data-mining techniques, where the typical domain is indeed often something like R^1000. -------------------------------- Date and time: Thursday 26th May 2011 at 15:00 Location: Room 124, School of Computer Science Title: Biasing the Evolution of Modular Neural Networks Speaker: Victor Landassuri (http://www.cs.bham.ac.uk/~vlm/) Institution: School of Computer Science (http://www.cs.bham.ac.uk/) Abstract: Neural networks exist with varying degrees of modularity ranging from pure modular networks characterized by disjoint partitions of hidden nodes with no communication between modules, to pure homogeneous networks with significant connections throughout. In between are apparently homogeneous networks that can be seen to have some degree of modularity if the hidden nodes are reorganized appropriately. In this talk, a modularity measure is presented and applied to the rearrangement of nodes to create modules in homogeneous networks, and that is used to improve the EPNet algorithm to evolve modular neural networks. Experimental results on a simple classification task confirm that the new modular EPNet algorithm does indeed lead to more modular networks than the classical EPNet algorithm, without compromising the performance on the given task.