Machine Learning for Developers

Limitations of early models

The model itself now has most of the elements of any normal neural model, but it had its own problems. After some years of fast development, the publication of the book Perceptrons by Minsky and Papert in 1969 prompted a stir in the field because of its main idea that perceptrons could only work on linearly separable problems, which were only a very tiny part of the problems that the practitioners thought were solvable. In a sense, the book suggested that perceptrons were verging on being useless, with the exception of the simplest classification tasks.

This new-found deficiency could be represented by the inability of the model to represent the XOR function, which is a Boolean function with an output of 1 when ...