skip to: page content | links on this page | site navigation | footer (site information)

Keith Neo Kian Seng

main page
machine learning | gaussian process | source codes | additional stuffs
publication lists | Systems Modelling Group publications
personal interests | my computer
contact page | location map
more about myself | guest book
other links

Source Codes

small logo

Source Codes Site

This site contains most of the source codes that have been developed for my research work in Gaussian processes. Some of which are standard MATLAB scripts, where others are fast algorithms developed in Mex-C (for use under MATLAB environment).

One of the main reason for choosing Mex-C instead of C code is that MATLAB is particularly useful and quick for building up mathematical formulation, especially when dealing with matrices. C codes are generally efficient for performing fast linear algebra operations, but typically takes a long time to code. So, if users can integrate the ease-of-use using MATLAB and let the C code perform fast mathematical operations, especially those involving "loops", this will typically some of these hardcore data-crunching operations, typically present in Gaussian process and some other mathematical formulas.

MATLAB Toolboxes

I have uploaded here some MATLAB scripts, which will be useful for those working in the regime of Gaussian process prior models. Currently, I am preparing a couple of simple MATLAB toolboxes but, for the moment, I will upload my first version of Gaussian Process Toolbox. As and when I see fits, there will be more to come.

I have developed some of the toolboxes for use in MATLAB when it comes to computation of explicit Gaussian process operations. Instead of just throwing scripts, which may be over hundreds of them, I will follow MATLAB's idea of classifying them as Toolboxes. Below are some of them:

  1. GAUSSIAN PROCESS TOOLBOX 1.0 (Released on March 31, 2006)
  2. GAUSSIAN PROCESS TOEPLITZ TOOLBOX 1.0 (Under Beta testing)
  3. GAUSSIAN PROCESS SCHUR TOOLBOX 1.0 (Under Alpha testing)
  4. GAUSSIAN PROCESS EXTENDED TOOLBOX 1.0 (Under proposal and consideration)

The Gaussian Process Toolbox 1.0 will provide the necessary functions and scripts to perform modelling of standard Gaussian process. The dataset will be constrained to N < 1,000 depending on the memory available in your system.

System Requirements for my Toolboxes These are the system requirements for running the Toolboxes:

  • Operating System: Windows 98/ME/2000/XP or Linux 2.6.x
  • Memory: 256MB (minimum) or 512MB (recommended to run MATLAB 7.0 and above)
  • Software:
    • MATLAB 6.5 and above
    • Optimization Toolbox 3.0 and above (from MathsWork Inc.)
    • Compatible C compiler recommended by MATLAB (i.e. MSVC and GCC)

Though I have listed the requirement for MATLAB is at least version 6.5 and above, the codes have been tested only on version 7.0. User may still experience bugs when running under version 6.5. In addition, Optimization Toolbox 2.1 may just be sufficient, but a later version if recommended. Please contact me if you have any of such encounter.

Background Information During the past few years, I have developed some source codes during the years of my research. Some of which are very useful in terms of developing fast algorithms for structured matrices.

Some of these codes include C codes and MATLAB codes for Gaussian process(Machine Learning). Fast algorithms developed under Mex-C codes were included. Instructions to compile Mex-C codes will be discussed here as well.

Thanks to several discussions with my supervisor, Professor W. E. Leithead and fellow colleague, Dr. Yunong Zhang, that have made it possible.
About Me | Site Map | Private Policy | Contact Me | ©2008 Keith Neo