Why “Building Blocks” Don’t Work on Parity Problems
Langdon, WB, Roberto Poli · UCL Discovery (University College London) · 1998
We investigate the distribution of performance of the parity and always-on Boolean functions given only the appropriate building block. These problems have “needle in a haystack” fitness landscape and so are unsuitable for genetic programming or other progressive search techniques. Theoretical analysis shows in the limit as program size grows the density of solutions is independent of size but falls exponentially with number of inputs.