Sonoma State University
CS Colloquium | April 6, 2006

The Self-Splitting Neural Network

Scott Gordon, California State University, Sacramento

Stevenson Hall 1300
12:00 PM - 12:50 PM

A Neural Network is a popular artificial intelligence technique in which an array of simulated neurons are interconnected to form a very simplified model of a tiny brain. Knowledge is stored in the interconnections, and is typically acquired through training. In practice, this is often limited to relatively small problems. The self-splitting neural network attempts to tackle larger problems by dividing the input domain into small chunks, and assigning a separate neural network to each chunk. Using a variety of splitting methods, large problems not typically solvable with standard neural networks have been learned quickly and with excellent generalization.