Waleed Kadous' PhD Thesis ... what's it all about?

As you probably know by now, I WAS doing a PhD at the University of New South Wales. My PhD is in machine learning in temporal domains. Machine Learning is a subfield of Artificial Intelligence, which is a subfield of Computer Science.

But now I am finished -- well, not exactly. I'm still waiting on the reviews to come back. They could refuse to give me a PhD, advise major or minor corrections.

You may also be interested in my Temporal Machine Learning page, which has links to researchers working on similar problems.

Here are some photos of the system: Equipment Glove In operation Everything

Publications and texts

• A less-technical description of my research. Being a researcher, even when I try to be non-technical, it doesn't always work. Still, it should be easy to read.
• My student abstract for Doctoral Consortium '98, which is part of AAAI '98, a conference I will be going to.
• Deaf Studies Research Symposium paper. This is a non-technical description, focusing on one of the domains I'm interested in -- Auslan. It's designed for a general audience and goes into very little technical detail. Also available in PostScript. (added 13 August 1998).
• Technical Report TR98-06: A General Architecture for Supervised Classification of Multivariate Time Series. Also available in gzipped postscript from here. Or see the abstract.
• ICML 99 paper.
  • Abstract
  • Paper (HTML)
  • Gzipped Postscript (about 360K).
  Note that it will probably change before finally getting published.
• Paper submitted to ICML2002:
  • HTML format.
  • Gzipped Postscript: about 80K.
  • PDF: about 215K.
• PhD Thesis (Updated 10 December 2002 -- new version with corrections recommended by examiners):
  • HTML format.
  • Gzipped Postscript: (about 1.5 megabyte).
  • PDF: about 1.8 megabytes.
• Updated 11 August 2003: TClass Source Code (the code used in my thesis) is available here.

Data Collected

Flock Sign Language Data

This is the data used in my thesis. It's available here as a bzipped, tarred, file. The file consists of 9 subdirectories tctodd1-9. Each directory consists of 3 samples of each sign, captured on a different day. In total there are 95 different signs, with 27 samples per sign. Signs were provided by a native signer volunteer.

Each file consists of a sequence of lines. Each line consists of 22 whitespace-separated numbers representing the 22 channels of information. The list of channels can be found in the domain description file. It also lists the classes. More information can be found in my PhD thesis above, in particular, section 6.3.1, and some in my honours thesis.

Finally, there's a list of the files and their class labels.

The files are:

• tctodd.tar.bz2 is the file with all the recordings of the signs.
• Domain description file lists the classes and the channels.
• Files and their class labels has a list of each file and its class label.