Robust Evolution Strategies

This paper empirically investigates the use and behaviour of Evolution Strategies (ES) algorithms on problems such as function optimisation and the use of evolutionary artificial neural networks in evolutionary robotics. Computer simulations are conducted which compare the performance of Classical-ES (CES) and Robust-ES (RES). We show that the performance of the RES algorithm improves on that of the CES algorithm. Most importantly statistical analyses of the evolutionary behaviour show that the CES algorithm keeps the same search strategy regardless of domain while the RES algorithm changes the search strategy to fit the problem.

• Type: Conference
• Year: 1998
• Where: SEAL
• Authors: Kazuhiro Ohkura, Yoshiyuki Matsumura, Kanji Ueda
• Keywords: Artificial Neural Networks, Evolutionary Robotics, Machine Learning, SEAL 1998, Search Strategy