Extended Curriculum Vitae: David Lindsay BSc. Jnt. Hons.

For a shorthand version of my CV (that fits onto 2 pages) please click on the following links to view the file in either word format , or pdf format .

Click on the bookmarks below to visit different sections of an extended version of my CV:

• Contact Details
• Profile
• Education and Qualifications
• Awards
• Skills
• Employment History
• Personal Details
• References
• Publications

Contact Details

50 Anders Corner,
Wokingham Road,
Bracknell,
Berkshire,
RG42 1PU
0771 4586880 / 01344 423747
d.g.lindsay@rhul.ac.uk

Profile

A self-motivated, honest and hard-working individual with an open-minded, committed approach to scientific research and its application to real life problems.

Education & Qualifications

PhD Student Machine Learning.
Computer Learning Research Centre, Royal Holloway, University of London, Surrey. 2002 - present.
This research investigated the quality of probability forecasts made by Machine Learning algorithms on highly complex real life data from various problem domains: finance, medical, bioinformatics, chemistry, physics, text mining and sports/gambling. I have introduced various meta-learning algorithms that output reliable probability forecasts, which are highly practical for cost-sensitive decision making domains. During my PhD I have been involved with teaching various undergraduate courses and was awarded the College Teaching Award For Postgraduates 2004 for my innovative teaching material.

Computer Science and Mathematics BSc. Jnt. Hons. (1st class) Overall average 87%. Royal Holloway, University of London, Egham, Surrey. 1999 - 2002.

'A' Levels

The Brakenhale School, Rectory Lane, Bracknell, Berkshire. 1996 - 1999.

GCSE's

The Brakenhale School, Bracknell, Berkshire. 1991 - 1996.

Wildridings Primary School, 1986 - 1991.

Awards

• College Teaching Award For Postgraduates 2004
• Best Computer Science Finalist 2002 Medal
• IMA Award 2002 Excellent Joint Mathematical Science Finalist
• Driver Prize 2001 First Year Essay on Fractals

Skills

• Training Courses: BCS Pattern Recognition Summer School (Exeter 2003), Training on the EPSRC funding process.
• Programming Skills: Java, MATLAB, Perl, Bash Shell, Latex, C++, Pascal, JavaScript, HTML, XML, CSS, SQL, Prolog.
• IT Skills: Microsoft Word, Access, Excel, Publisher, PowerPoint, Internet Explorer, Netscape, FrontPage, Outlook Adobe Photoshop, Lotus Softsuite, Page Maker, Paint Shop Pro, Dreamweaver, Hot Metal Pro, Visio, Windows .31, '95, '98, NT, XP, Unix/Linux.
• Communication & Social Skills: I developed these skills well during jobs where I was using the telephone and when I began teaching IT at my university.
• General work-related skills: These skills were acquired from every job I have had. These include time management, organisation, fire/health/safety training and so on.
• Other: Key skills communication level 3, purple belt in kickboxing, robotics design using LEGO mindstorms.
  

Employment History

Personal Details

References

Professor Volodya Vovk (PhD Supervisor), Room 111, Computer Science Dept., Royal Holloway University of London, Egham, Surrey, TW20 OEX. vovk@cs.rhul.ac.uk

Professor Glyn Harman, Room 244, Mathematics Dept., McRea Building, Royal Holloway University of London, Egham, Surrey TW20 OEX. g.harman@rhul.ac.uk

Chris Horton, Computer Centre, Royal Holloway University of London, Egham, Surrey, TW20 OEX. c.horton@rhul.ac.uk

Publications (With Brief Description)

Lindsay D., Cox S. Improving the Reliability of Decision Tree and Naive Bayes Learners, Fourth IEEE International Conference on Data Mining (ICDM 2004), Pages 459-462 November 2004

Description: Summary of the below technical reports, testing meta-learning implmentations of the Venn Probability Machine (VPM) and Binning techniques to impove the reliability of the probability forecasts output by the popularly used Naive Bayes and C4.5 Decision Tree Techniques. My Empirical Reliability Curve was also introduced in this paper as a useful visualisation of the reliability of probability forecasts.

Vovk V., Lindsay D., Nouretdinov I. and Gammerman A., Mondrian Confidence Machine, On-line Compression Modelling Project, http://vovk.net/kp, Working Paper #4

Description: This paper detailed various practically useful extensions to the Confidence Machine framework, inspired in part by the work of Piet Mondrian.

Lindsay D., Cox S.Effective Confidence Region Prediction Using Probability Forecasters, 10th Conference on Artificial Intelligence in Medicine (AIME 2005), To Appear July 2005.

Description: Work unifying the two practically useful extensions to pattern classification: Probability Forecasting (estimating conditional probabilities) and Confidence Region Prediction (narrowing down true labels with guaranteed probability of error). This work introduces a simple (yet provably valid) technique of converting probability forecasts into well-calibrated region predictions, and demonstrated this empirically on 15 publicly available datasets using various learning techniques.

Lindsay D., Cox S. Effective Probability Forecasting for Time Series Data Using Standard Machine Learning Techniques, 3rd International Conference on Advances in Pattern Recognition (ICAPR 2005), To Appear August 2005.

Description: Development of my research on reliable probability forecasting from traditional i.i.d. data to more complex time series data. Using a sliding window approach, I tested various machine learning methods: Naive Bayes, Decision Tree, Neural Network, SVM and Hidden Markov Models in an online learning setting (true labels given to learner with a delay) attempting to predict some discretised value in the time series sequence.

Lindsay.D., Visualising and improving reliability - a machine learning perspective. CLRC-TR-04-01, Technical Report, Computer Learning Research Centre, Royal Holloway University of London, Egham, Surrey, UK, 2004.

Description: A survey of the practical usefulness of probability forecasts output by commonly used machine learning techniques: K-Nearest Neighbours, Decision Tree, Naive Bayes, Bayesian Belief Networks, Neural Networks, Support Vector Machine (SVM). Also popular meta-learning techniques such as Binning, Boosting, Bagging and Find Best Weights are applied on top of the above learners to improve reliability of probability forecasts.

Lindsay. D., Reliable Probability Forecasting Using the Venn Probability Machine Learner. CLRC-TR-04-01, Technical Report, Computer Learning Research Centre, Royal Holloway University of London, Egham, Surrey, UK, 2004.

Description: Detailing my adaptation to the VPM meta-learning framework to generate probability forecasts. The VPM is applied on top of various learning algorithms: K-Nearest Neighbours, Decision Tree, Naive Bayes, Neural Networks and SVM.

Lindsay D., Effective Multi-Class Probability Forecasting with K29 Meta-Learners, 21st Conference on Uncertainty in Artificial Intelligence (UAI 2005), Under Review.

Description: The recently introduced Defensive Forecasting K29 algorithm is a kernel based approach to
probability forecasting which does not make any assumptions about the probability distribution underlying the data. This research details two extensions of the K29 Algorithm: 1) From binary to mult-class problems using a Pairwise Coupling methodology and 2) adaptation of the kernel technique as a meta-learning framework.

Lindsay D., Probabilistic Prediction of Sports Events: A Comparison Of Bookies and Machine Learning Methods, Fifth IEEE International Conference on Data Mining (ICDM 2005), Under Review.

Description: Experiments on Premiership Football and Flat Horse Racing data gathered from 2000 to 2005. A comparison of the performance of bookies starting price (converted into probability forecasts aiming to determine the outcome of the sporting event) and those made by various machine learning techniques and the recently introduced K29 algorithm are made. The results demonstrate a dramatic improvement over the bookies odds by the K29 learning algorithm, perhaps explained by the high dimensional analysis and lack of assumptions made on the data by the K29 approach.