Iterative competition is a catalyst for better problem solving

Kaggle is more than a machine learning competition platform; it’s a facilitator for efficient problem solving and a community for sharing and learning.

By Anthony Goldbloom
June 27, 2019
Race track Race track (source: annca via Pixabay)

Most people think of machine learning competitions in connection with Kaggle, but there’s a lot more to the platform. In this presentation, Kaggle founder and CEO Anthony Goldbloom talks about how the company grew from a few platform members to more than two million, and how and why it expanded beyond competitions.

Highlights from Goldbloom’s presentation include:

Learn faster. Dig deeper. See farther.

Join the O'Reilly online learning platform. Get a free trial today and find answers on the fly, or master something new and useful.

Learn more

Kaggle competitions show that iterative competition is a catalyst for better problem solving. Goldbloom notes that the live public leaderboard has a significant direct impact on performance, with competitors continually trying to one-up one another. “It’s a very efficient way to not draw a false conclusion on a problem,” Golbloom says. “And you know once you’ve run a competition, you’re at the limit of what’s possible.” (4:45)

The public rankings of competitors in the Kaggle competitions aren’t just good for bragging rights — the rankings are becoming a widely recognized machine learning credential. Goldbloom shares an example of Kaggle competitor “Giba,” who was ranked number 2 and in constant battle for first place: “ Giba, whose real name is Gilberto, is from Brazil. He was an electrical engineer at Petrobras. As a result of his performances on Kaggle, he got poached by Airbnb. And that is the case for a lot of the people in the top couple hundred on Kaggle.” (7:37)

The Kaggle team noticed that users were trying to share code and tutorials in their forum environments, which weren’t ideal (static code, missing libraries, and language version issues were just a few of the issues). So, Goldbloom explains, they launched Kaggle Kernels to facilitate sharing code, and then expanded to allow users to share full data sets, analysis related to — or even unrelated to — competitions, and to offer short tutorial courses. (10:46)

Post topics: AI & ML, Data, Next Economy
Post tags: Radar Event

Get the O’Reilly Radar Trends to Watch newsletter