Methods, Challenges and Successes of Machine learning prior to the 20th Century

As promised on my previous blog, let’s see what methods, challenges and successes happened in machine learning prior to 20th century.

Prior to the 20th century the idea of machine learning was mostly knowledge driven, with a vision to automate learning by these machines so that the knowledge could be passed to others, in a way that a human being is unable to.  This was a great idea, as we all know that people get old and retire or they leave companies, although they can do a handover, fact remains their knowledge and expertise always leaves with them.

Techniques like decision trees, neural networks, multi-layered networks were used in training machines. As with any subject of exploration machine learning was characterised by some challenges; those were

Difficulty to get a sufficient degree of randomness built into the structure. The expense of creating a device large enough to exhibit behaviour not significantly influenced by the operation of any one of its components. Slow response, theoretical limitations and not enough data to learn from.

Despite all these there were some instances of success that were reported like the use of chaostron by the U.S. Navy for controlling their inventory, application of decision trees to industrial process controls and the integration of explanation-based learning into general knowledge-intensive reasoning systems.

After all was there still potential for advancement in machine learning.  Let’s find out in our next episode.

References:

CADWALLADER-COHEN, J., ZYSICZK , W., & DONNELLY, R. (1984). THE CHAOSTRON: AN IMPORTANT ADVANCE IN LEARNING MACHINES. Communications of the ACM, 356-357.
Carbonell, J. G. (1989). Introduction: Paradigms for Machine Learning. Elsevier Science Publisher, 1-9.

Jones, R. M., & Taube, M. (1961). Notes on distinction between character recognition machines and percieving machines. American Documentations, 292.