Evolving Neural Feedforward Networks
نویسندگان
چکیده
For many practical problem domains the use of neural networks has led to very satisfactory results. Nevertheless the choice of an appropriate, problem specific network architecture still remains a very poorly understood task. Given an actual problem, one can choose a few different architectures, train the chosen architectures a few times and finally select the architecturewith the best behaviour. But, of course, there may exist totally different and much more suited topologies. In this paper we present a genetic algorithm driven network generator that evolves neural feedforward network architectures for specific problems. Our system ENZO optimizesboth thenetwork topologyand the connection weights at the same time, thereby saving an order of magnitude in necessary learning time. Together with our new concept to solve the crucial neural network problem of permuted internal representations this approach provides an efficient and successfull crossover operator. This makes ENZO very appropriate to manage the large networks needed in application oriented domains. In experiments with three different applications our system generated very successful networks. The generated topologies possess distinct improvements referring to network size, learning time, and generalization ability.
منابع مشابه
Feedforward neural network training using intelligent global harmony search
Harmony search algorithm is a meta-heuristic optimization method imitating the music improvisation process, where musicians improvise their instruments’ pitches searching for a perfect state of harmony. First, an improved harmony search algorithm is presented using the concept of swarm intelligence. Next, it is employed for training feedforward neural networks for three benchmark classification...
متن کاملEvolving the Architecture and Weights of Neural Networks Using a Weight Mapping Approach
The application of genetic programming to the evolution of neural networks has been hindered by the inadequacy of parse trees to represent oriented graphs, and by the lack of a good mechanism for encoding the weights. In this work, a hybrid method is introduced, where genetic programming evolves a mapping function to adapt the weights, whereas a genetic algorithm-based approach evolves the arch...
متن کاملEvolving Neural Networks for Classifier Prediction with XCSF
We extend XCS with computed prediction by replacing the usual linear prediction used in XCSF with a feedforward multilayer neural network. In XCSF with neural prediction, XCSFNN, classifier exploits a neural network to approximate the payoff surface associated to the target problem while the genetic algorithm adapts both classifier conditions, classifier actions, and the network structure. We c...
متن کاملEvolving Neural Networks for Hang Seng Stock Index Forecast
This paper describes an evolutionary neural network approach to Hang Seng stock index forecast. In this approach, a feedforward neural network is evolved using an evolutionary programming algorithm. Both the weights and architectures (i.e., connectivity of the network) are evolved in the same evolutionary process. The network may grow as well as shrink. The experimental results show that the ev...
متن کاملEvolving Spiking Networks for Turbulence-Tolerant Quadrotor Control
We investigate the automatic development of robust quadrotor neurocontrollers based on spiking neural networks. A self-adaptive evolutionary algorithm is used to generate highutility topology/weight combinations in the networks, and a simple synaptic plasticity mechanism provides some degree of in-trial adaptation. Incremental evolution gradually increases the severity of environmental conditio...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 1993