There is much to unlearn about computer and information
systems for understanding machine learning.
Traditionally, business intelligence using Information
processing systems has been about data warehousing, online analytical
processing, meta-data repositories. While all of them use past data and
experience for knowledge and discovery – there are differences. The one thing
in common is that they use sorting and searching algorithms with sequences of
instructions which lead to definite responses from the systems. Machine
learning is different.
Data mining models and applications come closest – except
they use very large data bases. Typically, data mining tasks are grouping
people or objects based on selected characteristics, estimating for inferring
or guessing reasonably, predicting expecting behaviors as a consequence of
previous actions. The list goes on with more statistical models with deeper
selective techniques for evaluating relationships and associations.
The talk about machine learning is around predictions using
Big data. A super market chain or Amazon
would prefer to know the customer’s next choice of items based on previous
purchases. Customer too prefers to get to the items he needs faster. These
recommendations and matching systems work with recognizing behavioral patterns
obtained from the data. What is derived from the pattern recognition is a response
of likelihood of the behavior. Detecting frauds, stock market predictions with
changing data and environment require learning and intelligence more than just depending
on the system design using a large database. Applications recognizing speech
and faces, robotics, biometrics use machine learning techniques. Machine
learning broadly encompasses supervised learning (through association,
regression, classification) and unsupervised learning for pattern recognition –
an engineering term for classifying.
Questions to ponder:
What can go right and wrong with machine learning systems?
What work is suitable for learning machines in platforms?
How does machine learning contribute to Platform Value (its
feedback system, market places, data orchestration, network effects) ?
References:
Loshin, David. Business Intelligence, The Savvy Manager’s
guide, Morgam Kauffman Publishers, 2003.
Alpaydin, Ethem. Introduction to Machine Learning, edited by Ethem
Alpaydin, MIT Press, 2014.