There are two major approaches to machine intelligence and pattern recognition. The first tries to remake the world according to the mind of the computer by creating a clear, logical representation of the pattern. The second tries to fit the data with a statistical representation. Neither is good for both cases, but both have their successes.
Logical Matrix
Many patterns in the world make logical sense. Day follows night. Spring follows winter. Logical pattern-recognition algorithms are great at finding these patterns whether they are big or small.
There are two parts to these algorithms: a pattern generator and a pattern evaluator. The algorithms generate plausible patterns and test them against th
e data to see if they fit it well. The evaluator must choose the best pattern that is often the simplest.
Statistical Whims
But many corners of the world are not cut perfectly. How old is old? When is soon? Problems like these demand guesstimates that statistical algorithms generate well. These algorithms compare the sequences against each other and effectively render an opinion. Correlation matrices and covariance are important to making the final decision.
Combining the two algorithms often makes the most sense. A logic algorithm can define a pattern that comes close, and a statistical algorithm can refine the parameters.
One of these things is not like the other, and that algorithm in the corner is going to tell us which.
Patterns In The Code
