Memory Based Reasoning

An mechanic who fixes an engine by recalling another car 

that exhibited similar symptoms,

A lawyer who advocates a particular outcome in a trial based

on legal precedents of like cases,

An engineer who copies working elements of nature (practicing biomimicry),

by treating nature as a database of solutions to problems,

So, what is common between them?

They are all solving new problems based on the solutions of similar past problems.

From the stables of Roger Schank and his students at Yale University in the early 1980s, came a pioneering technique called Case Based Reasoning,

Its generalized form is a two step approach - Identify similar cases from experience and apply the information from these cases to the problem at hand,

This empirical classification method is also called as memory-based reasoning (MBR),

In the statistical world, its application lies in finding cases/records in a database that are similar to a specified pattern,

MBR finds neighbors similar to a new record and uses those neighbors for classification and prediction,

This operation requires two key functions – a distance function that assigns a distance between any two records and a combination function that combines the results from the neighbors to arrive at an answer,

The distance function, combination function, and number of neighbors are the key ingredients in determining how good MBR is at producing results.