Learning Statistically Neutral Tasks without Expert Guidance
Ton Weijters, Antal P. J. van den Bosch, Eric O. Postma · 1999
In this paper, we question the necessity of levels of expert-guided abstraction in learning hard, statistically neutral classification tasks. We focus on two tasks, date calculation and parity-12, that are claimed to require intermediate levels of abstraction that must be defined by a human expert. We challenge this claim by demonstrating empirically that a single hidden-layer bp-som network can learn both tasks without guidance. Moreover, we analyze the network 's solution for the parity-12 task and show that its solution makes use of an elegant intermediary checksum computation. 1 Introduction Breaking up a complex task into many smaller and simpler subtasks facilitates its solution. Such task decomposition has proved to be a successful technique in developing algorithms and in building theories of cognition. In their study and modeling of the human problem-solving process, Newell and Simon [1] employed protocol analysis to determine the subtasks human subjects employ in so...