is a naturally inspired artificial intelligence Algorithm. GNPvs learns from data using biological evolutionary processes such as mutation and crossover. As a Network structure, GNPvs can learn problems from simple to complex and is able to adapt to the complexity by itself. This can be reached because, in contrast to many other AI algorithms, GNPvs can grow and shrink in size.
GNPvs is an artificial intelligence that allows learning patterns and rules from data. By variable tuning the network structures, a GNPvs automatically adapts to the complexity of a given problem. Furthermore, the GNPvs selects suitable features by itself. Unlike artificial neural networks, the decision-making process of GNPvs can be replicated and visualized at any time. Therefore, the GNPvs is not a black box model.
The downloadable thesis on the right presents an extension of the GNP by two novel mutation operators. These novel operators allow the GNP to be variable in the number of nodes and to change the function outputs of nodes. Extending the search space without danger concerning overfitting or the bloat problem is feasible using both operators. With a higher solution capability, it is now possible for the GNP to automatically adapt to the complexity of a given task and to find suitable features of high dimensional data sets. The mutation operators are successfully applied to a financial data set where there could improve the standard GNP and show better performance concerning portfolio optimization. Experiments showed that the standard GNP could be beaten by 61.84% return on the financial data.
The video on the left shows the evolution of the best individual of a GNPvs trained by the Iris dataset. The initial network consisting of only three nodes evolves into a sufficiently complex network to perform an excellent classification. This is possible by adding nodes to a GNPvs. The left half of the video shows the genotype, and the right half shows the transition path.
The video on the right shows the evolution of the best individual of a GNPvs trained by the Iris dataset. The initial network consists of 40 judgment nodes and 40 processing nodes. Such a complex network cannot initially describe the simple complexity of the problem well. However, the GNPvs can produce excellent results by deleting nodes from the network. The left half of the video shows the genotype, and the right half the transition path.