It can be argued that soft computation methods rely heavily on inductive reasoning and thus can suffer from weaknesses introduced by the problem of induction. Explain how these problems might manifest in the following contexts:
a) The process by which a single layer perceptron is trained to categorize patterns; and
b) Flaws inherent in the use of fully-trained perceptrons to categorize patterns.