Modern Machine Learning – Overview with simple examples

Modern Machine Learning – Overview with simple examples

Introduction – A machine of any kind takes time to be accustomed of, which can be termed as the learning curve for both the machine and its operator and the troubles faced by the operator during this time are called as the teething problems. During this stage, while the operator gets used to the nuances of practical usage of the machine, the machine also gets used to the practical applications of itself.

History – In the digital world machine learning is quite different than the physical world. In 80s and 90s and till the first decade of 21st century machine learning was often associated with the cognitive learning and almost always implemented  using very sophisticated algorithms through neural network or at least AI models. (As late as in the mid of nineties Fuzzy Logic had created quite a stir in this area.) At that time, the PC evolution was at a very nascent stage and the computing power of the computers was very limited and therefore the high-end algorithms envisaged during those days, either used to remain unimplemented or were tried using the few supercomputers developed across the world.

Evolution – With the advent of Internet and wireless technologies and the revolution in RAM storage capacities, there was a sudden surge in the computational power of the computers and now we see complex phenomena like voice and image analysis made very simple. With this, the machine learning, which, once upon a time was only for the sophisticated developers arena, has become everyone’s buzzword.

However, considering the market and business orientation of the technology, the contemporary machine learning has been limited to data and pattern only and not high-end cognitive learning as imagined earlier. Even the most sophisticated robots of today’s world don’t go into weaving dreams or genetic analysis and their maximum application remains within the weather forecasting and aerodynamic calculation and in the apparel and entertainment industry.

In the contemporary view, the machine learning is of the following kinds as described below. (Certain desired action follows the mapping.)

Case:

Objective – Yearly report to be produced in multiple threads (a single thread would take longer)

Supervised learning –   Provide the thread identifiers (month names, department names, customer names, geography names etc) and let the program constitute the data for each thread.

Semi-supervised learning – Let the program analyze the data in each column and come up with the thread identifiers and then constitute the data for each thread. The risk here is, unless any guidance given for the thread size, the threads could be of dissimilar in size defeating the purpose.

Unsupervised learning– Let the program analyze the optimum thread size and then analyze the data in each column and come up with the thread identifiers and then constitute the data for each thread.

Examples of Machine learning in SAP EAM environment:

Serialization

Supervised – Serialization based on given criteria

Semi-Supervised – Serialization based on past experience

Order and notification type determination

Supervised – Flags supplied as part of the data from the interfaces

Active – Based on the device acting as data source

Compatible units

Supervised – Manual mapping of equipments

Semi supervised – Automatic mapping of equipments