Open Access Open Access  Restricted Access Subscription Access

An Overview of Artificial Neural Networks: Part 4 Learning Mechanism


   Subscribe/Renew Journal


This paper presents the concepts of Learning Mechanism in Artificial Neural Networks (ANNs). The set of exact rules for the resolution of a learning difficulty is called a learning algorithm. Every learning algorithm varies from the other in the approach in which the modification to a synaptic weight of a neuron is expressed. There are different learning rules; several of them are discussed in this paper. Hebbian, Delta, Competitive, Memory based, Outstar and Boltzmann learning rules are discussed in detail, tabulated and compared in terms of weight adjustment, initial weight setting and different learning pragdism i.e. supervised or unsupervised. The learning rules are deliberated with consequence, its separate mathematical justification and applicability. Definitely, ANNs can be trained using these rules to perform meaningful tasks such as grouping, recognition or relationship. The concept of Perceptron is given discussed in next part of this paper series.


Keywords

Hebbian Learning, Delta Learning Rule, Competitive Learning Rule, Memory Based Learning, Outstar Learning Rule Boltzmann Learning Rule.
User
Subscription Login to verify subscription
Notifications
Font Size


  • An Overview of Artificial Neural Networks: Part 4 Learning Mechanism

Abstract Views: 945  | 

Authors

Abstract


This paper presents the concepts of Learning Mechanism in Artificial Neural Networks (ANNs). The set of exact rules for the resolution of a learning difficulty is called a learning algorithm. Every learning algorithm varies from the other in the approach in which the modification to a synaptic weight of a neuron is expressed. There are different learning rules; several of them are discussed in this paper. Hebbian, Delta, Competitive, Memory based, Outstar and Boltzmann learning rules are discussed in detail, tabulated and compared in terms of weight adjustment, initial weight setting and different learning pragdism i.e. supervised or unsupervised. The learning rules are deliberated with consequence, its separate mathematical justification and applicability. Definitely, ANNs can be trained using these rules to perform meaningful tasks such as grouping, recognition or relationship. The concept of Perceptron is given discussed in next part of this paper series.


Keywords


Hebbian Learning, Delta Learning Rule, Competitive Learning Rule, Memory Based Learning, Outstar Learning Rule Boltzmann Learning Rule.

References