114 M.F. Korns
The results support the view that the pursuit if high accuracy algorithms in noiseless
training data also conveys distinct and measurable advantages with noisy training
data and range shifted testing data. In fact, for both noiseless training data and when
trained on noisy training data, then tested on range shifted testing data, the enhanced
EA algorithm is measurably faster and more accurate than the baseline algorithm.
This places the Extreme Accuracy algorithm in a class by itself.
The new EA algorithm introduces a hybrid view of SR in which advanced
evolutionary methods are deployed in the extremely large spaces where serial search
is impractical, and in which the intractable smaller spaces are first identified and
then attacked either serially or with mathematical treatments. All academics and SR
researchers are heartily invited into this newly openedplayground, as a plethora of
intellectual work awaits. Increasing SR’s demonstrable range of extreme accuracy
will require that new intractable subspaces be identified and that new mathematical
treatments be devised.
Future research must explore the possibility of developing an Extreme Accuracy
algorithm for the related field of symbolic multinomial classification.
Finally, to the extent that the reasoning in thisinformal argument,ofextreme
accuracy, gain academic and commercial acceptance, a climate ofbeliefin SR can
be created wherein SR is increasingly seen as a “must have” tool in the scientific
arsenal.
Truly knowing the strength’s and weaknesses of our tools is an essential step in
gaining trust in their use.
About the Author
Michael Kornsis a computer scientist with professional experience at IBM
Research, Chief Scientist at Tymeshare Transactions, Chief Scientist at Xerox
Imaging, and currently CEO of Korns Associates. His primary area of interest
is in symbolic regression classification with an emphasis on investment finance
applications. His most recent research has explored extreme accuracy algorithms
for symbolic regression which are robust even in the face of noisy training data and
range shifted testing data.
References
Hornby GS (2006) Age-layered population structure for reducing the problem of premature
convergence. In: GECCO 2006: Proceedings of the 8th annual conference on genetic and
evolutionary computation. ACM, New York
Korns M (2010) Abstract expression grammar symbolic regression. In: Genetic programming
theory and practice VIII. Springer, New York, Kaufmann Publishers, San Francisco, CA
Korns M (2011) Accuracy in symbolic regression. In: Genetic programming theory and practice
IX. Springer, New York, Kaufmann Publishers, San Francisco CA