| DEPARTMENT
OF
COMPUTING SCIENCE AND MATHEMATICS |
|
|
|
| UNIVERSITY . COMPUTING SCIENCE . SEMINARS | ||
SEMINARS - Autumn 2003
The Department of Computing Science and Mathematics presents the following seminars. Unless otherwise stated, seminars will take place in Room 4B94 of the Cottrell Building, University of Stirling from 15.00 to 16.00 on Friday afternoons during semester time.
If you would like to give a seminar to the department in future or if you need more information, please contact the seminar organisers, either David Cairns (Phone 01786 467445, Email dec@cs.stir.ac.uk) or Julie Hodgkin (Phone 01786 467446, Email jho@cs.stir.ac.uk).
Talk Schedule
[Top] [Abstracts]
26th September 2003 [Schedule]
Combating Infinite State using Ergo3rd October 2003 [Schedule]Carron Shankland
Symbolic transition systems can be used to represent infinite state systems in a finite manner. The modal logic FULL, defined over symbolic transition systems, allows properties over infinite state to be expressed, establishing necessary constraints on data. We present here a theory and tactics for FULL, developed using Ergo, a generalised sequent calculus style theorem prover allowing interactive proofs. This allows exploitation of the underlying symbolic transition system and reasoning about symbolic values.
This is joint work with Peter Robinson, University of Queensland and will be presented at FORTE 2003 next week.
Javier Marin-BlazquezHyper-Heuristics: Learning to combine simple heuristics
The idea underlying hyper-heuristics is to discover some combination of familiar, straightforward heuristics that performs very well across a whole range of problems. To be worthwhile, such a combination should outperform all of the constituent heuristics. In the talk I will describe a couple of approaches (a learning classifier system and a messy-GA-based approach) that learn such a heuristic combination for solving one-dimensional bin-packing problems. Some ongoing work on Timetabling problems will be outlined as well. Current hyper-heuristic research is focused on constructive approaches (the final solution is built from scratch step by step). When applied to a large set of bin-packing benchmark problems, the learned procedure finds an optimal solution for nearly 80% of them, and for the rest produces an answer very close to optimal. When compared with its own constituent heuristics, it ranks first in 98% of the problems.
Dynamics of storage and recall in associative memories: what can we learn from cortical control structures?17th October 2003 [Schedule]Bruce Graham
Associative memory is arguably the fundamental process by which information is stored and recalled in the brain. It is one of the longest-standing applications of artificial neural networks. Though it has been of immense theoretical interest, real-world computational devices that incorporate an associative memory are rare. Problems here include restrictions on the format of information that can be stored, and the rapid deterioration in memory performance once capacity is exceeded in ANN implementations of associative memory. Though such models are likened to brain operation, particularly in the mammalian hippocampus, they capture little of the architecture and dynamic operation of brain circuits. The continuous accumulation of experimental data from the mammalian hippocampus provides a wealth of information from which to try to glean a better understanding of associative memory operation in the brain. Neurons receive spatially and temporally distributed inputs that interact with a complex milieu of intrinsic membrane conductances to generate and modulate neuronal output and plasticity. Neuronal types include excitatory pyramidal cells and a diverse range of inhibitory interneurons. These neuronal types are connected in specific network architectures, with the interneuronal networks providing the dynamic control system for the operation of the excitatory cell networks. In this talk I will introduce the idea of associative memory and how it has been implemented by artificial neural networks. I will then consider the circuitry and dynamics of the mammalian hippocampus. The similarities and differences with the ANN models will be highlighted. A key issue is how storage and recall may be phased in the brain.
Challenges of Computer Mathematics24th October 2003 [Schedule]Henk Barendregt
Systems of computer algebra can deal with symbolic computations better than humans. In general these systems represent "computable" mathematical objects. At present systems are being developed that can represent on a computer arbitrary mathematical objects, computable or not. The long range goal is that these proof-assistants will help human users to develop mathematical theories. We will discuss how these systems work, how far one has succeeded in representing mathematics and why the enterprise is interesting already at present.
Models of Context: Theory, Systems, and Scalability7th November 2003 [Schedule]Paddy Nixon
The first part of the talk will describe the Global Smart Spaces (GLOSS) project. We will outline the core developments in models of context and their implementation. The models are role based, component oriented models and provide a compositional framework for the device and service discovery in pervasive systems. The focus will be on software aspects and issues of scalability in the model. We will conclude the talk by presenting intuitions and early ideas for a theortical foundation for this work.
Image features grouping for stereopsisNicolas Pugeault
Stereopsis allows us to recover depth information on a scene from a pair of video sensors. The main difficulty is to associate features in one image with corresponding ones in the other. In previous work, a number of local processings have been proposed to address this correspondence problem. A purely local matching does not require global assumption over the scene being witnessed, yet the limited amount of information makes this process ambiguous and susceptible to noise. In previous work, Norbert has shown that using conjointly, modalities improve the quality and stability of stereo matching. Here I propose to use higher order events, involving the neighbours of a local features, to discriminate more accurately between locally ambiguous choices. A feature is associated to its neighbours depending on their proximity and good continuation, in our multimodal formulation. I will show how consideration of constellations of local primitives can improve our stereopsis.
14th November 2003 [Schedule]
Electronic Plagiarism Detection Software
Lynne Blair
This seminar provides an opportunity to find out more about electronic plagiarism detection software. We will look primarily at Turnitin and CopyCatch, but will also mention the software JPlag, for detecting source code plagiarism. “Electronic detection can only be an adjunct to the normal exercise of academic judgement” Jude Carroll & Jon Appleton, "Use of electronic detection tools", http://www.jiscpas.ac.uk/site/guide7.asp
Google has been cited in the past as being the most commonly used tool to help to confirm plagiarism in written text, perhaps because of its familiarity and ease of use. However, this relies on the "intelligent guessing" of phrases likely to be copied and, if you guess wrong, it can take huge amounts of time to see this through.
Whereas certain pieces of text may exhibit an obvious "change of style" at some point, perhaps indicating a different author, in general it is far from obvious or easy to pick the right phrases. Many electronic plagiarism detection systems now exist to help with this process, and their use is now starting to grow significantly, partially due to the recent activities of the JISC Plagiarism Advisory Service. Some systems are designed primarily for checking free-text against internet sources (e.g. Turnitin) whilst others are designed primarily to check for collusion between students (e.g. CopyCatch). There are also further systems that are designed to check for plagiarism in computer programs for a range of programming languages (e.g. JPlag and MOSS).
Ongoing studies at Lancaster are looking at further integrating such systems into our assessment process across the University. Our initial judgement of systems was based largely on a series of 4 JISC surveys/ reports of various electronic detection systems (JISC Electronic Plagiarism Detection project, http://www.jisc.ac.uk/index.cfm?name=programme_plagiarism). A summary of the systems in use at Lancaster is given below. This seminar provides an opportunity to find out more on the operation of these systems, particularly Turnitin and CopyCatch.
Turnitin – for powerful web searches (exact string matching) - JISC operated, avoids legal concerns with using US system - Free for academic institutions for 2 years (from Sept '02) - High administrative overhead w.r.t. data protection, copyright (e.g. student permissions) - http://www.turnitin.com or http://www.jiscpas.ac.uk
CopyCatch – for powerful collusion checks (incl. paraphrasing) - UK-based commercial operation (linguistic foundation) - License required for full version (£250 single user, £375 2 users - Oct '02) - Cut-down version now available free for academic institutions - Minimal administrative overhead - http://www.copycatch.freeserve.co.uk
JPlag – for source code plagiarism detection - German based software; free for academic institutions - http://www.jplag.de
21st November 2003 [Schedule]
Low Level Driver Development Under Linux
Bernd Porr
I'm developing a device for data acquisition under Linux. For this device I also had to write a device driver. In this talk I'll present some aspects of the driver development under Linux. This will include: the module concept of drivers, how drivers communicate with other drivers, how drivers access the hardware and how the the userland communicates with them. In particular I will show how to write a new driver for the USB subsystem. Finally I'll point out problems with the actual driver model and show some solutions.
28th November 2003 [Schedule]
Model 2X: a better way of Web publishing?Previous Seminar Series [Top] [Abstracts] [Schedule]Richard Bland
Many authors have suggested that an excellent way of publishing Web material is to use Java servlets within a MVC (Model-View-Controller) architecture. Julien Mercay and Gilbert Bouzeid have proposed a particular implementation strategy for this idea, which they call Model 2X. Their strategy uses XML and XSLT within Jakarta Struts. The seminar will review Model 2X and discuss other ways of implementing the general approach.
2003 - Spring2000 - Autumn
Last Modified: 6th March 2003