Computing Science Seminars, Autumn 2014
The Division of Computing Science and Mathematics presents the following seminars. Unless otherwise stated, seminars will take place in Room 2A73 of the Cottrell Building, University of Stirling from 15.00 to 16.00 on Friday afternoons during semester time. For instructions on how to get to the University, please look at the following routes.
Autumn 2014
| Date | Presenter | Title/Abstract |
|---|---|---|
| Friday 26 Sep |
Prof
John McCall Professor of Computer Science Director of IDEAS Research Institute Robert Gordon University of Aberdeen |
Telling the Wood from the
Trees: Essential Structure in Model-based Search.
Problem structure, or linkage, refers to the interaction
between variables in an optimisation problem. Discovering
structure is a feature of a range of search algorithms
that use structural models at each iteration to determine
the trajectory of the search. Examples include Information
Geometry Optimisation (IGO), Covariance Matrix Adaptation
Evolution Strategy (CMA-ES), Bayesian Evolutionary
Learning (BEL) and Estimation of Distribution Algorithms
(EDA) In particular, EDAs use probabilistic graphical models to represent structure learned from evaluated solutions. Various EDA approaches using trees and graphs have been developed and evaluated on a range of benchmarks. Benchmarks typically have "known problem structure" determined a priori. In practice, the relationship between the problem structure, structures found by by the EDA and algorithm success is a complex one. This talk will explore these ideas using a classification of problems based on monotonicity invariance. We completely classify all functions on 3 bits and show that conventional concepts of the relationship between structure and problem difficulty are too brittle to capture the subtlety even of this low dimensional function space. The talk concludes with a discussion of related work on algorithm classification and some speculative ideas for topological notions of essential problem structure. |
| Friday 3 Oct |
Dr
Savi Maharaj Lecturer in Computing Science Computing Science and Mathematics University of Stirling |
Using Computers to Model
and Study Human Behaviour - First Steps. The choices
we make and the way we behave have important consequences
both for us as individuals and for society. Computational
techniques can help us both to model and to study human
behaviour through techniques such as agent-based modeling
and virtual experiments. I have been using both of these
approaches in my work. As human behaviour is of course a
huge topic, this talk will not even attempt to cover it
comprehensively, but will focus on two (possibly three)
projects. These involve modelling and investigating
behaviour in the contexts of epidemic spread and
environmental protection policy, and are collaborations
with psychologists and/or economists. The talk will cover
the kinds of models used, the experimental techniques used
to study real human behaviour in order to parameterise the
models, some tentative results, and the difficulties
encountered in inducing (and recognising) realistic
behaviour in virtual experiments. The talk will end with a
speculative discussion of the way forward, with a surprise
bonus contribution from students Craig Docherty and Daniel
Gibbs. |
| Friday 10 Oct |
Alasdair
P Anderson Chief Architect, Big Data for HSBC Group, HSBC BANK PLC |
Data Monetization:
Leveraging Legacy with Big Data. HSBC is a large
international company, with about 250,000 staff and lots
of data (about 65Pbytes across over 7000 systems).
They need a single globalised platform, rapidly, with
advanced analytical capabilities. How can they go about
getting this? |
| Friday 17 Oct |
Dr
Iain J Gallagher Lecturer in Health & Exercise School of Sport University of Stirling |
Big Data Analysis in
Biology & Sport. Big data techniques have been
routinely used in biology for a number of years. These
large data sets are gathered using 'omics' technologies
and allow biologists to examine almost all of the entities
that make cells and tissues function at a particular
moment in time. This informs on important differences in
the molecular details of diseased and normal tissue, in
the processes and pathways active during development or
disease and in the small molecular differences that
predispose us as individuals to disease. Thus as well as
providing mechanistic insight big data in biology is
beginning to allow the stratification of individuals by
disease risk. In sport big data is also becoming more and
more important. Applications include prediction of player
performance and propensity for injury, real time object
detection as well as the analysis of results and betting
patterns (both legal and illegal). In this talk I will
give an overview of big data and the analysis of these
data in both biology and sport. After introducing the
concepts of big data in these domains I will use a 'toy',
but real dataset of Olympic heptathlon results to
demonstrate how dimensional reduction can help us identify
athletic characteristics that could be used for
competitive advantage. I will then give a brief overview
of the 'omics' technologies currently used in biology to
gather large amounts of information – snapshots of
cellular or tissue activity. I will briefly discuss my
tools of choice for big data analysis and the concept of
'reproducible research'. Finally, time permitting I will
present a couple of examples briefly illustrating how a
programmatic approach can inform biological questions. |
| Friday 24 Oct |
Dr Rachel
Lintott Research Assistant Computing Science and Mathematics University of Stirling |
Using Process Algebra to Model Radiation Induced Bystander Effects. Radiation induced bystander effects are secondary effects caused by the production of chemical signals by cells in response to radiation. I will present a Bio-PEPA model which builds on previous modelling work in this field to predict: the surviving fraction of cells in response to radiation, the relative proportion of cell death caused by bystander signalling, the risk of non-lethal damage and the probability of observing bystander signalling for a given dose. This work provides the foundation for modelling bystander effects caused by biologically realistic dose distributions, with implications for cancer therapies. |
| Friday 31 Oct |
Mid Semester | Mid Semester |
| Friday 7 Nov |
Dr
Nanlin Jin Lecturer Computer Science and Digital Technologies Northumbria University, Newcastle |
Smart Meter Data Analytics
by Subgroup Discovery. Based on a recent article.
This talk will cover the following three aspects:
|
| Friday 14 Nov |
Dr Alasdair
Rutherford Lecturer in Quantitative Methods School of Applied Social Science University of Stirling |
Working with and analysing
administrative data from linked health and social care
records in Scotland. Better evidence is
needed on how social care and health care interact to
enable people to live independently in their own homes for
as long as possible. In most countries, individual-level
administrative data on social care, housing support and
health care are collected separately by different
organisations. Linking such datasets could enable service
providers, planners and policy makers to gain an improved
understanding of how these services can work together more
effectively to improve outcomes for patients and service
users. In collaboration with the Scottish Government’s
Health Analytical Services Division and the Information
Services Division of NHS Scotland we have analysed a new
dataset produced as a pilot project which brings together
detailed information on individual social care packages,
prescribing, diagnoses (including dementia and mental
health) and hospital episodes. This enables us to explore
pathways through health and social care from the
individual perspective. This presentation will describe
the building of the dataset, challenges in preparing and
analysing the data, and provide some initial findings from
the study. |
| Friday 21 Nov Room: 4B108 |
Dr Kira
Mourao Research Associate School of Informatics University of Edinburgh |
Learning Relational Action
Models in Robot Worlds. When a robot, dialog
manager or other agent operates autonomously in a
real-world domain, it uses a model of the dynamics of its
domain to plan its actions. Typically, these domain models
are prespecified by a human designer and then used by AI
planners to generate plans. However, creating domain
models is notoriously difficult. Furthermore, to be truly
autonomous, agents must be able to learn their own models
of world dynamics. An alternative therefore is to learn
domain models from observations, either via known
successful plans or through exploration of the world. This
route is also challenging, as agents often do not operate
in a perfect world: both actions and observations may be
unreliable. In this talk I will present a method which, unlike other approaches, can learn from both observed successful plans and from action traces generated by exploration. Importantly, the method is robust in a variety of settings, able to learn useful domain models when observations are noisy and incomplete, or when action effects are noisy or non-deterministic. The approach first builds a classification model to predict the effects of actions, and then derives explicit planning operators from the classifiers. Through a range of experiments using International Planning Competition domains and a real robot domain, I will show that this approach learns accurate domain models suitable for use by standard planners. I also demonstrate that where settings are comparable, the results equal or surpass the performance of state-of-the-art methods. |
| Friday 28 Nov |
John Page Senior Solutions Architect MongoDB Glasgow |
MongoDB, a Big data solution for the real world. This talk will introduce the concepts behind MongoDB - the world's most prevalent and popular big data solution and leading NoSQL Database, explaining what matters when you have a 'Big Data' problem in a real world application and how many of the world's leading companies address this. |
| Friday 5 Dic |
Patricia
Ryser-Welch Department of Electronics The University of York |
Can Plug-and-Play hyper-heuristic help
create state-of-the-art human-readable algorithms?
Hyper-heuristics are defined as heuristics optimising
heuristics. Previous approaches focus on the quality of
the numerical solutions obtained; very little discussion
concentrates on the generated algorithms themselves. Some
hyper-heuristic frameworks tend to be highly constrained
and their architecture prevents the search from
finding state-of-the-art algorithms. Often the
generated algorithms are also not human-readable. We
discuss a new interpretation of hyper-heuristic
techniques, that explicitly and openly offers a flexible
and modular architecture. This together with analysis
of evolved algorithms, could lead to
human-competitive results. |
Top image: A simplified view
of the neural circuit of the CA1 region of the mammalian
hippocampus (Fig. 2 pg 284 "Hippocampal Microcircuits"
book, coeditored by Prof. Bruce Graham,
Springer 2010). Encoding and retrieval of patterns of
information in this circuit have been studied by computer
simulation, as detailed in Cutsuridis, Cobb & Graham,
Hippocampus 20:423-446, 2010.
Courtesy of Prof.
Bruce Graham
Last Updated: 25 November 2014.