Constructive Neural Network Learning Algorithms

Constructive Algorithms offer an approach for incremental construction of potentially minimal neural network architectures for pattern classification tasks. These algorithms obviate the need for an ad-hoc apriori choice of the network topology. The constructive algorithm design involves alternately augmenting the existing network topology by adding one or more threshold logic units and training the newly added threshold neuron(s) using a stable variant of the perceptron learning algorithm (e.g., pocket algorithm, thermal perceptron, and barycentric correction procedure). Several constructive algorithms including tower, pyramid, tiling, upstart, and perceptron cascade have been proposed for a-category pattern classification. These algorithms differ in terms of their topological and connectivity constraints as well as the training strategies used for individual neurons.