HOW do you improve the performance of a simple neural network? Use a tank of water to train it, say computational neurobiologists who believe the discovery may help us understand more about how the brain processes information.
The experiment carried out by Chrisantha Fernando and Sampsa Sojakka at the University of Sussex, UK, involved a type of neural net known as a perceptron, which can be “trained” by example to produce a specific output in response to a number of different inputs.
There are well-known limits to what a perceptron can be trained to do. For example, a perceptron can…
