Inference Group

Research activity report, January 2000 - David MacKay

Error-correcting codes

IBM Research labs, Zurich, have given us a $40,000 award with no strings attached. They are also interested in supporting a PhD student who would work in Zurich for x months per year, where x is being negotiated. I hope to have this student start in September 2000.

Lucent (Bell labs) have developed a chip for use in disc drives based on the error correcting codes that we have been working on. I am asking Lucent for support similar to IBM's.

I am still talking to Nortel, and to another company called Tandberg TV, about industrial collaboration.

Matthew Davey defended his thesis in December 1999. We have submitted a paper on a method for communicating over channels with synchronization errors, a long-neglected research area.

Simon Wilson now says he will submit his thesis by February 2000.

Latent variable models

James Miskin continues to work on blind signal separation, including deblurring of images. He is working on data from various sources, including Toshiba's Cambridge laboratory.

Human-computer interfaces

David Ward is improving the adaptive language models for our gesture-driven computer interface, Dasher. He has made a prototype coupled to an automated dictation system so that experimental tests can easily be made. Alan Blackwell, a postdoc in the Rainbow Group at the Computer Laboratory, has joined the project, and has helped us to assess the useability of the interface. Depending on the outcome of these experiments, we may try to spin off the software into handheld computers and/or computers for disabled people.

Once the finger-driven version of the interface is working well, we will continue work on the eye-driven version using the eye-tracker that has been loaned to us by ASL.

Teaching Physics

Sanjoy Mahajan continues to develop Physics teaching methods; his test groups include high school students, first years and second years. He collaborates with Mazur and Hestenes in the USA.

We plan to create hands-on Physics exhibits (`plors', as found in the Exploratorium) and test their effectiveness as teaching aids in local schools.

Other topics

Empirical Modelling of non-linear relationships
Harry Bhadeshia of Materials Science continues to use our neural network and Gaussian process software.
Quantum computing
The new PhD student who was to work on quantum error-correction did not come.

Plans for next year

I hope to recruit two PhD students, one funded by IBM, and one supported through the Mott system; and perhaps one postdoc.
David MacKay <>
Last modified: Fri Jan 7 21:46:07 2000