Solving a Non-Linear Optimization problem…to win Jeopardy!
It was a few years ago now, but Laura McLay recently reminded me of the story of Roger Craig, all-time record holder for Jeopardy! who used a mixture of Big Data analysis and simulation to (very successfully) prepare for the show.
You can find all of the details in the book Final Jeopardy: Man vs. Machine and the Quest to Know Everything, but in a nutshell, Dr. Craig is remembered not only for his record breaking wins, but for how he trained himself for the show. He extensively studied the past questions maintained on the J! Archive website, parsing the whole website to create a large, unstructured data set. He then identified which categories to study based on their value in the game and his perceived and tested knowledge of the categories. The algorithm was based on the probability of getting questions correct for his “predicted self” using simulation. He then worked to optimize his now quantified knowledge for the predicted game.
As a published academic, Roger does a much better job of explaining it than I do so watch his talk at ‘Quantified Self’ in New York as shown on YouTube: