×

Optimized artificial neural networks

  • US 6,601,053 B1
  • Filed: 05/25/2000
  • Issued: 07/29/2003
  • Est. Priority Date: 05/19/1989
  • Status: Expired due to Fees
First Claim
Patent Images

1. A method of producing optimized artificial neural networks, comprising:

  • providing a plurality of different neural network input-output sets including a plurality of training sets and an evaluation set;

    providing a population of multiple symbol strings, the symbols of the strings representing parameters of neural network architecture and parameters of a learning method;

    providing artificial neural networks for respective symbol strings of the population, based on the neural network architecture parameters;

    training the artificial neural networks using the learning method depending on the training sets and learning method parameters;

    evaluating the trained neural networks using the evaluation set to generate evaluations of the trained neural networks;

    generating a further population of multiple symbol strings from the population of symbol strings using a genetic algorithm depending on the evaluations;

    continuing the training, evaluation, and generating until a stopping criteria is met; and

    selecting one of the symbol strings for producing the optimized artificial neural network depending on the evaluations.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×