Photo: University of Michigan
Data prediction startup Kaggle has helped NASA map dark matter, helped Allstate predict car accidents, and let Heritage Health Prize predict who will go to the hospital next year.Now it’s tackling a less sexy, but equally important problem: automatically grading high school essays, so tired teachers don’t have to.
The Hewlett Foundation is sponsoring a competition: the goal is for somebody on the Kaggle platform to get as close as possible to predicting the already marked grade of 23,000 high school student essays.
The current leader in the competition is 28-year-old Martin O’Leary, a research fellow in glaciology at the University of Michigan.
Translation: He predicts how glaciers move for a living!
Anthony Goldbloom, co-founder of Kaggle, told us how O’Leary cracked the code in a previous Kaggle competition to map dark matter and he did this in just a week. Although he ended up finishing fourth, O’Leary’s efforts didn’t go unrecognised.
O’Leary tweeted about the experience: “Not braggin’ or nothin’ but the White House just compared me to Newton and Einstein.”
Goldbloom told us: “Many of our members are very smart people, but are not stimulated at work. Rather than doing a Sudoku puzzle, they get to solve very valuable problems. What’s interesting is the diversity of the people who are performing well. You’d expect people with a deep understanding in education to do well, but the people are totally random.”
Like for example:
Vik Paruchuri is a financial developer at VCP Management. He’s in second place.
Marchin Pionnier is a 32-year-old programmer from Poland. He’s third right now.
Momchil Georgiev is fourth, and he’s a programmer for the National Weather Service.
There’s a team that is currently in fifth place. They competed in previous Kaggle competitions separately, but decided to band together for this one. The five-person team is diverse, ranging from an engineer for AT&T to a bioinformatics Ph.D. at Rutgers University to a consultant at Deloitte.
The competition ends April 23, 2012. The top three members will get a piece of the $100,000 prize. Right now, there are only 25 teams, so the odds are good!
“The winning algorithm will likely be acquired by companies that sell existing scoring engines,” Goldbloom said.
Goldbloom was talking at the Founders Den tonight as part of a demo day for startups there.