process one command at a time. With current processor architectures this is only
partially true due to multiple pipelines with parallel execution of commands,
but the algorithms are usually not designed to take full advantage of parallel
execution. The main motivation for designing classifier algorithms of parallel
nature are to increase the speed of execution and to potentially handle larger
classification problems. Examples of parallel classifier algorithms include the
SPRINT decision-tree-based classifier by Shafer et al. [1996a] and the Kerneltron
hardware Support Vector Machine implementation by Genov and Cauwenberghs
[2003].
tion approaches.
classifier algorithm is used there exists at least one data distribution where ran-
dom guessing is better. Or in other words: even the "most fancy" nonlinear
classifier is not the best for all occasions.