Further reading

Valiant introduced the distribution-free PAC criterion [ Valiant , 1984 ] and later formulated evolvability as learning by local mutation and selection driven by aggregate empirical fitness rather than by individual examples [ Valiant , 2009 ] . Vapnik and Chervonenkis developed the uniform convergence and shattering framework underlying statistical control of infinite classes [ Vapnik and Chervonenkis , 1971 ] ; the relation among learnability, VC dimension, and finite-sample bounds was sharpened by Blumer et al. [ 1989 ] . Gold’s identification in the limit supplies the contrasting exact, asymptotic presentation paradigm [ Gold , 1967 ] . Chapters 3 and 7 develop its ORACLE and UOCL forms; the present chapter supplies their query-relative statistical relaxation.