The magazine archive includes every article published in Communications of the ACM for over the past 50 years.

Interest in machine learning can be traced back to the early days of computer science. Alan Turing himself conjectured that some form of automatic learning would be required to endow a computer with artificial intelligence. The development of machine learning, however, has not always lived up to these ambitious beginnings. The first flood of interest in machine learning was generated by Rosenblatt's perceptron. 4 The perceptron is a simple thresholded linear classifier that can be trained from an online sequence of examples. Intriguingly, from a theoretical point of view, the number of mistakes during training can be bounded in terms of intuitive geometric properties that measure a problem's difficulty. 3 One such property is the margin, or the distance of the most" borderline" example to the perceptron's decision boundary. First highlighted by the mistake bound in perceptrons, the margin would come to play an important role in the subsequent development of machine learning; however, its importance in Rosenblatt's time was not fully realized. Interest in perceptrons waned after a critical study by Minsky and Papert that catalogued their limitations. 2