Antibodies teach computers to learn
- 06 January 1996
- Magazine issue 2011
FIRST came neural networks, then genetic algorithms. Now it's the turn of software inspired by the human immune system. Researchers at the University of Wales in Aberystwyth have turned to the immune system because it is remarkably good at learning.
To protect the body from disease, the immune system must learn to recognise foreign substances, or antigens, and keep them on record for very long periods of time. The body does this by generating a variety of antibodies that bind to various antigens and destroy them. If a familiar antigen turns up, it is instantly recognised for what it is and neutralised. If an antibody gets the better of an antigen, it clones itself to increase the number of similar antibodies in the general population. The cloning process sometimes introduces random mutations, which may make the antibody better or worse at tackling the problem.
Immunity to a particular antigen fades ...