Robotics projects

Contents Page

Publications

 

Artificial Immune Systems

AINE, RLAIS and SSAIS

I first got involved in AIS research when I took over supervision of Jon Timmis’ PhD. He came up with an algorithm using a network structure based on Jerne’s idiotypic network theory. The algorithm was a bit half-baked, but worked reasonably well for some data sets. It wasn’t terribly easy to get a good structure out of without some fiddling. It was “nearly-one-shot” learning, which meant that you had to be careful to do just the right number of passes through your data set to avoid having a network that was either wildly too big or too connected.

 

So we moved on to a variety of other algorithms that used the idea of limitation of resources to moderate network growth. We stuck at it for a few years. After a couple of years of not working (I went sailing) I came up with a slightly different way of building the networks. It now does what I always wanted it to do. The new algorithm does:

 

  • Form stable memories, even in the absence of stimulation by patterns similar to those being remembered
  • Continuous learning: you can go on training the networks forever without detrimental effect
  • Locally distributes resources and does not have centralised control. This helps to make it computationally efficient.

 

Some rough and ready software that implements this algorithm and a simple network visualizer are available (see link below).

 

This is the first (probably buggy/incomplete) version of this package that I have released. Feel free to use for non-profit activities, make alterations, analyse data etc… If you write papers using or modifying this then please send me copies and make sure to cite this web-page and the papers detailed in the package. There are other more up to date publications (eg. proceedings of ICARIS 2003).

 

If you want to make lots of money from this, or develop the algorithm/software into something to sell then ask me first please. If you make any money, then I’ll want a cut…

 

Download from here: smainpack (about 5.5Mb).

 

Enjoy!

 

 

 

A network visualization of Fisher’s famous iris data.