Abstract
A logic for a committee of perceptrons, called seniority logic, and a local adjustment algorithm for training a seniority committee are described. Like a majority committee, a seniority committee has the capacity to solve any two-class problem in which the classes are disjoint. Unlike a majority committee, a seniority committee may have members added during training, and a seniority committee is free to attain a size needed to solve a problem. In computer simulations on patterns with binary components a seniority committee has a higher recognition rate and needs fewer members than a majority committee. Solutions learned during training have high recognition rates on the training set and are reasonably free of bias.

This publication has 1 reference indexed in Scilit: