Pattern Recognition Problem (from Baseball Musings)

In two of my past jobs, the people I worked with were interested various problems that could be classified as pattern recognition. Basically, a computer program is given a series of inputs, and what those inputs mean. From that, the computer program builds a probability model. Given an unknown input, the program uses the probability model to decide which of the items it knows about matches the input best.

The example all of you know about is Google. You give the Google search engine a pattern of words, and Google gives you a list of files that best match your query. If you give the same inputs to a different search engine, you get different results. One reason is that different search engines cover different parts of the web. The other is that they use different algorithms to find the answer.

Humans are really good at this. Hitters in baseball do this all the time, recognizing a pitch from the arm angle of the pitcher, his release point, the spin of the ball, and many other variables they don't even know they are using. However, not all inputs are always available to us, so we make do with what we have and what we trust. The strikeout + error last night is a great example of this.

There were three decisions made last night based on different inputs from the play. Josh Paul, the catcher, used a single input in determining that the batter was out; he caught the ball. In his experience, that was enough to tell him the batter was out. The fielders on the Angels, and the fans in the stands just had a visual input. All game long they saw the swinging third strike call by Eddings. When they saw the exact same call on Pierzynski their model told them the batter was out. That was their signal the inning was over, and the Angels ran off the field.

Pierzynski, however, was in a unique position to get the play right. He caught the whole game with his back to the umpire. That forced A.J. to build a probability model based on auditory, not visual signals. So when he struck out and headed to the dugout, his brain set off an alert. The auditory model he had built listening to Eddings was screaming, "You're not out!" Because of his experience making the same mistake as Paul, Pierzynski believed the auditory model and headed to first. A.J. was the only person involved who made the right decision. Any other White Sox hitter would not have built the correct pattern recognizer, and the game would have been decided in extra innings.