Acknowledgements

My thanks go firstly, as they should always be, to The Unique God (whom Muslims call Allah), the one who blessed me with the ability to undertake and finally complete this work.

My family, of course, deserve great thanks for their immense support, as does my wife. They have been patient, encouraging and understanding. I thank my father for his advice, my mother for her guidance, and my siblings Hediah, Kareema, Hassan and Amatullah for their tolerance. My wife, Agnes Chong, also deserves my unending gratitude for her help on innumerable fronts.

My gratitude also extends to my teachers and staff at the University of New South Wales: Claude Sammut for his supervision, Andrew Taylor for his encouragement, Ross Quinlan for his assistance and advice especially in the early stages of the PhD, Arun Sharma for keeping my nose to the grindstone and the staff the Computing Support Group for their assistance with hardware and operating system related issues.

I also owe a great debt to my local colleagues: Andrew Mitchell for helping me think through some of the issues in the PhD, Phil Preston for being an excellent sounding board on new ideas and a guide on all things practical, Charles Willock for his presentation of the alternative point of view on so many issues, Mark Reid for helping with lots of things, but particularly the hairy maths, Paul Wong for his friendship and advice and Michael Harries for his advice. Thanks also to Ashesh Mahidadia for pointing out that metafeatures may have applications beyond temporal classification. I would also like to thank Bernhard Hengst and Mark Peters for sharing their experiences and for proofreading early drafts; and also to two other proofreaders: Peter Gammie and Peter Rickwood.

My thanks also extend to those who helped with the datasets: with the Sign Language dataset, my thanks go particularly to Todd Wright, but also to those who helped with the undergraduate work: Adam Schembri and Adam Young. Also Philip de Chazal and Branko Celler for providing me with the ECG data, not to mention Philip's thesis which contained a wealth of resources. And Peter Vamplew for many discussions and data on automated recognition of sign language.

And also to my international colleagues: Eamonn Keogh for advice and discussion of his work, Hiroshi Motoda for his assistance with graph-based induction, Tom Dietterich for his advice and encouragement, to Ivan Bratko for his assistance; and also to the many interesting people I met at conferences and with whom I had many interesting conversations.

Finally to the Open Source and Free Software communities for provision of many excellent tools - in particular I would like to thank everyone involved in Weka, especially Eibe Frank and Ian Witten.

I'm sure I've forgotten someone. I assure you that this is a shortcoming on my part and not on yours. I beg you to forgive me for my oversight.

Mathematical terminology

Symbol	Meaning
$\langle a, b, c \rangle$	A tuple (in this case triple) of values
$\{a, b, c\}$	A set with three elements $a, b, c$
$\{a_1, ..., a_n\}$	A set containing the elements $a_1$, $a_2$ etc.
	all the way to $a_n$
$a \in A$	The value $a$ is an element of $A$
$[a, b, c]$	A list containing the elements $a, b, c$
$\{x\vert b(x)\}$	The set of all values of $x$ for which $b(x)$ is true
$f: A \rightarrow B$	$f$ is a function that maps from an element
	of $A$ into an element of $B$
$\mathrm{domain}(f)$	If $f: A \rightarrow B$, then $\mathrm{domain}(f)=A$
$\mathrm{range}(f)$	If $f: A \rightarrow B$, then $\mathrm{range}(f)=B$
$\mathbb{P}(X)$	The set of all possible subsets of $X$
$X^*$	The set of all lists generated
	by selection of elements from $X$
$X^+$	$X^* - []$
$e_{\langle i \rangle}$	The $i$th value of the tuple $e$
$\sum_{s \in S}f(s)$	The sum of $f(s)$ over all elements in $S$
$\sum_{i=m}^{n}f(i)$	$f(m) + f(m+1) + ... + f(n)$
$\mathop{\rm argmin}_{s \in S} f(s)$	The value of $s$ for which $f(s)$ is the least
$\mathop{\rm argmax}_{s \in S} f(s)$	The value of $s$ for which $f(s)$ is the most

Abbreviations

Abbrev	Meaning
Abbrev	Abbreviation
AI	Artificial Intelligence
Auslan	Australian Sign Language
CBF	Cylinder-Bell-Funnel
DTW	Dynamic Time Warping
ECG	Electrocardiograph
FIR	Finite Impulse Response
HMM	Hidden Markov Model
HTK	HMM Tool Kit
ILP	Inductive Logic Programming
ML	Machine Learning
PEP	Parametrised Event Primitive
RNN	Recurrent Neural Network
SMI	Signal Match Image
TC	Temporal Classification