netJEN

Evolution of Activation Functions for Artificial Neural Networks

The netJEN framework providing rich functionality for the evolution of ANNs should be improved and extended with the evolution of activation functions (AFs). In addition to the existing evolution of AFs, where AFs are selected from a given set of commonly used functions, evolution of arbitrary AFs based on concepts already studied with the netGEN system should be implemented. The shape of the AFs is determined by a set of evolved points connected by cubic splines. The netJEN GUI has to be extended with corresponding parameter settings and visualization of the evolved AFs.

Project State

In progress...

Project Supervisor

Helmut Mayer

Project Student

Required Knowledge

Java Programming
Artificial Neural Network Basics
Evolutionary Computation Basics

Acquired Knowledge

Java Swing
Java Frameworks
Evolution of Activation Functions
Spline Approximations

Contacts

netJEN System
August Mayer (Boone)

Financial Support

None

Project Duration

October '03 ---