PARITY IS NOT A GENERALIZATION PROBLEM

Uninformed learning mechanisms will not discover ''type-2'' regulariti es in their inputs, except fortuitously. Clark & Thornton argue that e rror back-propagation only learns the classical party problem - which is ''always pure type-2'' - because of restrictive assumptions implici t in the learning algorithm and network employed. Empirical analysis s howing that back-propagation fails to generalise on the party problem is cited to support their position. The reason for failure, however, i s that generalisation is simply not a relevant issue. Nothing can be g leaned about back-propagation in particular, or learning in general, f rom this failure.