Learning programs with an easy to calculate set of errors
William I. Gasarch, Ramesh K. Sitaraman, Carl H. Smith, Mahendran Velauthapillai · 1988
We continued the study of learning an approximation to the desired function. Rather than measure the variance between the desired function and the approximation, we accounted for the difficulty of deciding membership in the set points comprising the variance. Our results indicate that the more complex a decision procedure is allowed, the larger the class of functions that become inferrible.