Abstract
Goal is to build hyper-elastic 3-D models using statc
experiments. Focuses on discovering local energy
function using GP.
Splits deep evolution (ie crossover and mutation) from
surface evolution (parameter adjustment using random
process like Evolution Strategies). Uses generational
GPQuick. Crossover 0.3, mutation 0.5 and copy 0.2
Prepared to run to 1000 generations. Pop size
500.
Fitness based on closeness of match both of GP function
and also its derivative. (tentativly this seems to
increase both accuracy and robustness) However results
indicate GP is overfitting (due to single fitness
case?)
Users
Please
log in to take part in the discussion (add own reviews or comments).