Daisy, Daisy; May 1993; Scientific American Magazine; by Philip Yam; 2 Page(s)

What does a computer do when it starts to die? The HAL 9000 in the film 2001: A Space Odyssey burst into a rendition of "A Bicycle Built for Two," a song it had been taught early in life. The memorable scene may not be too far off the mark. That's what one researcher found out when he began to "kill" a type of computer program known as an artificial neural network. As the network approached death, it began to output not gibberish but information it had previously learned--its silicon life flashed before its eyes, so to speak.

The analogy to a so-called near-death experience is irresistible because the creators of artificial neural networks design them to mimic the structure and function of the biological brain. A neural network relies on "units" to serve as the cell body of a neuron and "links" between the units to act as the interconnecting dendrites and axons. The units are typically organized into several layers. A consequence of such an architecture is that the network, like the brain, can learn. In a real brain, learning is thought to occur because of changes in the strength of synaptic connections among neurons. Similarly, a neural network alters the strength of the links (specifically, the weighting between units) to produce the correct output. Typically a programmer teaches a network by repeatedly presenting training patterns to it [see "How Neural Networks Learn from Experience," by Geoffrey E. Hinton; SCIENTIFIC AMERICAN, September 1992].