Computer Science · Computational Theory and Mathematics
and
Hal Wasserman, Manuel Blum, Name Manuel Blum
See on the mapAbstract
We review the eld of result-checking, discussing simple checkers and self-correctors. We argue that such checkers could pro tably be incorporated in software as an aid to e cient debugging and enhanced reliability. We consider how to modify traditional checking methodologies to make them more appropriate for use in real-time, real-number computer systems. In particular, we suggest that checkers should be allowed to use stored randomness: i.e., that they should be allowed to generate, preprocess, and store random bits prior to run-time, and then to use this information repeatedly in a series of run-time checks. In a case study of checking ageneral real-number linear transformation (for example, a Fourier Transform), we present a simple checker which uses stored randomness, and a self-corrector which is particularly e cient if stored randomness is employed.