Guiding principle for Goal 2.
From http://devolab.cse.msu.edu/projects/
Meta-EC: Configuring Evolutionary Computation via Evolutionary Computation
One of the major promises of evolutionary computation (EC) is having computers solve difficult problems with minimal human intervention. In reality, however, getting EC to provide solutions to problems requires an extreme amount of arcane knowledge about how to choose a satisfactory setup of evolutionary parameters among the enormous number of possible configurations. This is due to the fact that what constitutes a good EC setup -- which genetic operators to use and with what frequency -- changes from problem to problem such that a poor configuration will fail to yield a valuable result. Given that EC itself is good at finding satisfactory solutions amongst large multi-dimensional search spaces, it makes sense to try to find good EC setups using EC. We explore the viability of this strategy, which we call "meta-EC" and preliminarily results indicate that it is a good means of automating the process of selecting parameter settings for EC. If successful, meta-EC research could help EC deliver its original promise.
Researchers: Clune, Goings, Goodman and Punch
Contact: jclune@msu.edu
Meta-EC: Configuring Evolutionary Computation via Evolutionary Computation
One of the major promises of evolutionary computation (EC) is having computers solve difficult problems with minimal human intervention. In reality, however, getting EC to provide solutions to problems requires an extreme amount of arcane knowledge about how to choose a satisfactory setup of evolutionary parameters among the enormous number of possible configurations. This is due to the fact that what constitutes a good EC setup -- which genetic operators to use and with what frequency -- changes from problem to problem such that a poor configuration will fail to yield a valuable result. Given that EC itself is good at finding satisfactory solutions amongst large multi-dimensional search spaces, it makes sense to try to find good EC setups using EC. We explore the viability of this strategy, which we call "meta-EC" and preliminarily results indicate that it is a good means of automating the process of selecting parameter settings for EC. If successful, meta-EC research could help EC deliver its original promise.
Researchers: Clune, Goings, Goodman and Punch
Contact: jclune@msu.edu