News Picks : Learning algorithm developed to mine large data sets
03 April 2014
Ars Technica: To analyze the explosion of data being generated in a range of scientific fields, an algorithm has been developed that works with minimal human supervision. Called KODAMA (knowledge discovery by accuracy maximization), it comprises five steps, one of which is iterative, to classify and evaluate the data and look for meaningful patterns. To test its accuracy, the developers have applied the algorithm to several data sets, including the linguistic characteristics of State of the Union addresses. In that exercise, KODAMA was able to identify a transition in vocabulary that occurred during Ronald Reagan’s presidency.