GENERAL AREAS: Comp. Vision, Neural Networks, Machine Learning,
AI, Pattern Recognition
APPLICATIONS: image segmentation, character recognition, shape classification, gesture recognition, intelligent web browsing, man-machine interfacing, fusion of information, various prediction and diagnostic tasks in different domains (inc. medical, financial)
PREREQUISITIES: Machine Intelligence, Pattern Recognition, Neural Networks, AI , Image Proceesing all help
P1: MACHINE LEARNING-Decision Trees mcs02.pdf
1) To implement an inductive machine learning algorithm that learns decision trees from error-free examples using concept of information measure
2) To investigate generalisation performance with different stopping criteria for synthesising decision trees of appropriate size when handling noisy data.
3) Improve performance by using pruning strategies to build smaller
trees
P2: Binary Labelling and Multiple Classifier Systems if01.pdf
1) To write an algorithm that uses constructive approach on binary data
2) Test the algorithm using benchmark data
3) Investigate the effect of different criteria for determining non-separability
of data
P3: SHAPE CLASSIFICATION gesture.html
1) Become experienced with NN package that implements Back-propagation.
2) Investigate various representations of 2D shape e.g hand gestures suitable for describing contours as input to a neural net.
3) Characterise robustness of simple shape classification by training
with data artificially corrupted with noise.
P4: RADIAL BASIS FUNCTION NETWORKS
1) Become experienced with NN package that implements RBF networks
2) Investigate feature extraction properties of RBF networks using various learning strategies
3) Use features derived from an RBF network to improve performance
of a multi-layer perceptron network
P5: Combining Multiple Classifiers ugprod.html
To implement a known combining technique (Bagging) with both a decision
tree and feedforward neural network as the base classifier, and to characterise
the advantages and disadvantages on a variety of classification tasks.
P6: Machine Learning Benchmark Data
To set up a suite of (existing) machine learning data sets from various
domains, and to produce a data set adviser which describes the examples
and suggests methods to solve them.
P7: Computer Aided Learning
To devise examples for use with an interactive program that teaches
basic principles
e.g. BackProp and RBF networks, Control System Design, Prolog, AI
P8: Neural networks for face recognition and detection cvpr01.pdf
P9: Error-Correcting codes and Multiple Classifier Systems kluwer02.pdf
P10: Bagging, Boosting and diversity measures
Other areas for student-suggested projects: Control, Multimedia Graphics, Visualisation
Dr Terry Windeatt, Centre for Vision, Speech and Sig. Proc., School of EE, IT and Maths
University of Surrey, Guildford, Surrey GU2 5XH
Phone: +44 (01483) 259286 Email: T.Windeatt@surrey.ac.uk