Pattern recognition in any data set

People have a natural ability to find patterns in arrays of information. In any seemingly meaningless set of numbers, words, actions or people, we are able to find some kind of system or even create (invent) it. The phenomenon of automatic formation of hierarchies and other patterns in society is well known to sociologists. How our brain does this is completely incomprehensible, which is why we still cannot create computer systems that have the same universal ability to analyze data. On the contrary, each data mining system has to be programmed to search for a certain, clearly described type of patterns.

Specialists in cognitive science and psychology from the Massachusetts Institute of Technology declare that for the first time in the world they have created a universal tool of this kind. Their program is able to recognize absolutely any patterns in the same way as the human brain does. At the input, it receives absolutely any stream of unstructured data, and then it shows which construction is most suitable for organizing this data — hierarchy, linear order, tree, cluster, or something else.

All our life and all nature is based on clear classifications: from the periodic system of chemical elements to the tree of evolution and social networks. Therefore, AI must learn to work with such objects.

After receiving the data set, the program structures them in the way that is most suitable for this case. This is exactly how our brain approaches the decision: we keep in mind all possible patterns and compare them with the set of numbers before our eyes. Actually, such abilities are one of the main measures when passing tests for intelligence.
')
MIT research results are published in the latest issue of Proceedings of the National Academy of Science. This development can be the foundation for a set of specialized systems that will work in the same bundle and recognize patterns of any kind, forming the core of artificial intelligence systems of the new generation.