Kaggle

What Kaggle is

Kaggle is a data science competition platform and an online community for data scientists and machine learning practitioners^[1]. Organisations post a problem together with data; participants build models, submit predictions, and are ranked against one another. The platform is also used to search and analyse public datasets, build machine learning models and grow data science expertise^[2][3].

How a competition runs

Most competition submissions are scored immediately against a hidden solution file and summarised on a live leaderboard^[1]. A participant therefore gets an answer within seconds rather than waiting for a review, and everyone can see where they stand at any moment. That combination of instant feedback and a public ranking is what turns a modelling exercise into a contest.

Origin

Kaggle was founded by Anthony Goldbloom in April 2010^[1]. Jeremy Howard, one of Kaggle's first users, joined in November 2010 as President and Chief Scientist, and Nicholas Gruen served as founding chair^[1]. In 2011 the company raised 12.5 million dollars and Max Levchin became chairman^[1].

What has been run on it

Notable competitions hosted on Kaggle include gesture recognition for Microsoft Kinect, a football AI for Manchester City, and improving the search for the Higgs boson at CERN^[1]. The subjects range from consumer hardware and sport to fundamental physics, and the same format — a dataset, a metric, a leaderboard — is used for all of them. Wikipedia notes that Geoffrey Hinton and George Dahl used deep neural networks to win a competition hosted by Merck^[1], an early example of a contest drawing attention to a particular modelling approach.

Joining Google

On 8 March 2017 Fei-Fei Li, Chief Scientist at Google, announced that Google was acquiring Kaggle^[1]. The Google Cloud Platform Blog post titled Welcome Kaggle to Google Cloud is dated Wednesday 8 March 2017 and is authored by Fei-Fei Li, Chief Scientist, Google Cloud AI and Machine Learning^[3]. In that post she wrote that Kaggle, founded in 2010, was home to the world's largest community of data scientists and machine learning enthusiasts, with more than 800,000 data experts using it^[2][3].

The post describes Kaggle as the best place to search and analyse public datasets, build machine learning models and grow data science expertise^[2][3]. Li wrote that with Kaggle joining the Google Cloud team, the companies could accelerate the mission of democratising AI and lowering barriers to entry^[3]. The post said that Kaggle and Google Cloud would continue to support machine learning training and deployment services while offering the community the ability to store and query large datasets^[3]. 9to5Google quotes Kaggle's own announcement that the team would remain together and continue Kaggle as a distinct brand within Google Cloud, continuing to grow its competitions and open data platforms^[4].

Growth

In June 2017 Kaggle surpassed 1 million registered users, and by October 2023 it had over 15 million users in 194 countries^[1]. Those numbers describe a community that grew by an order of magnitude after the acquisition, while the format itself — competitions scored on a live leaderboard, alongside public datasets and openly shared work — stayed the same.

Why it matters

Kaggle sits at the meeting point of three things: organisations with data and a hard problem, people who want to practise machine learning on real data, and a public record of which methods actually did well. A leaderboard makes results comparable in a way that a paper's reported number cannot easily be, and immediate scoring means a participant can iterate many times in a day. For learners it is a place to work with real datasets and see how their models compare; for the field it is a running archive of problems, submissions and rankings that shows what machine learning could do at a given moment.