Training Neural Networks Using Computational Verb Rules

To design efficient learning algorithms for neuralnetworks based on different kinds of experiences, of which someare static and some are dynamical, is one of the best ways tocope with the complexities of training neural networks. Thetraining process of a neural network is in general dynamical.Computational verb rules can be efficiently applied to improvethe training procedures of neural networks. First, the dynamicalexperiences of training neural networks were represented ascomputational verb rules. Then computational verb inferencesbased on computational verb similarities were used to improvethe training performances by tuning the learning constants.The methods of tuning learning constants were based on thedynamical experiences of the correlation between the convergentspeed of neural networks and the dynamics of learning constants.Some examples were presented to show the usefulness of thetraining algorithm provided.