Home ยป Project groups

REVAC

The REVAC method provides an information theoretic measure on how sensitive a parameter is to the choice of its value. This can be used to estimate the relevance of parameters, to choose between different possible sets of parameters, and to allocate resources to the calibration of relevant parameters. The method calibrates the parameters of a model to fulfill the given constraints while retaining a maximum of robustness and generalizability. While it was originally designed to select and calibrate the parameters of an evolutionary process, it can in fact be applied to any complex modelling task.

Project Index

The REVAC method provides an information theoretic measure on how sensitive a parameter is to the choice of its value. This can be used to estimate the relevance of parameters, to choose between different possible sets of parameters, and to allocate resources to the calibration of relevant parameters.

Measuring relevance by Shannon entropy

Because REVAC is implemented as a sequence of distributions with slowly decreasing Shannon entropy we can use the Shannon entropy of these distributions to estimate the minimum amount of information needed to reach a target performance level. We can also measure how this information is distributed over the parameters, resulting in a straightforward measure for parameter relevance.

Algorithm details

REVAC starts from an initial table TeX Embedding failed! that was drawn from the uniform distribution over TeX Embedding failed!. The update process that creates a new table TeX Embedding failed! from a given TeX Embedding failed! can be described from both the horizontal and the vertical perspective. Looking at Table from the horizontal perspective we can identify two basic steps:

Demonstration

This demonstration is based on the file calibrate_ga.m in the software bundle on the software page. The function calibrates a generational genetic algorithm (GA) on one of four test functions: sphere (f1), saddle (f2), step (f3), and Schaffer's f6 (f6).

Publications

Volker Nannen, S. K. Smit, and A. E. Eiben, Costs and Benefits of Tuning Parameters of Evolutionary Algorithms, Proceedings of the 20th Conference on Parallel Problem Solving from Nature, 2008. [pdf]

Software

A zip file with the matlab code used for the IJCAI-07 paper can be downloaded here. After unpacking you have a directory revac_demo/ with the following files and directories:

calibrate_ga.m A function that allows you to test REVAC on a genetic algorithm

Algorithm

REVAC uses information theory to measure parameter relevance. Instead of estimating the performance of an EA for specific parameter values or ranges of values, REVAC estimates the expected performance when parameter values are chosen from specific probability density distributions.