Google And Kaggle Hosting Video Understanding Contest

February 17, 2017 - Written By Daniel Fuller

Google has put out a call for help in improving YouTube’s video recognition and understanding algorithms in the form of a contest, held jointly with data science website Kaggle. Google is asking participants to help improve YouTube by building out a new machine learning model to help identify and analyze some 700,000 odd sample videos. The kicker is that they’re offering participants credits for Google Cloud Platform so that they can use it to host, build out, and power their machine learning models. Participants will also have access to Google’s open-source TensorFlow framework to give their machine learning exploits a boost with the power of neural networking and Google’s specialized TPU computing units. The top players in the contest will get to split a prize pool of $100,000. Contest winners will be honored at the upcoming IEEE Conference on Computer Vision and Pattern Recognition, and will give a presentation at the show.

YouTube has a ton of different things going on under the surface that help it to be as intuitive and feature-rich as it is, and one of those things is thousands of algorithms powered by machine learning that help to sort the various videos by their categories and content. Those algorithms do expand on their own, that being the nature of machine learning, but they do it slowly. YouTube staff help to expand the algorithms both by feeding the machines swaths of pre-sorted data, and by expanding the algorithms’ codebases manually. The data set that powers the whole show is called YouTube-8M, and this is what contest entrants would be helping to expand and analyze.

The contest at hand is basically giving civilian coders the chance to do what YouTube engineers do with largely similar tools. Google Cloud Platform and TensorFlow are largely the same tools that Google uses in-house, which means that the only thing that the coder would have to provide is a personal computer to hook into the whole arrangement, and their skills. Essentially, it’s a chance to get a taste of how Googlers work, with a $100,000 prize pool thrown in to sweeten the deal and ensure that coders who give up their valuable time have a shot at being handsomely compensated for it, as long as they provide value to Google first. If you know a thing or two about machine learning, are curious about the work side of life at Google, or want a chance to try out Google Cloud Platform and TensorFlow, you can head through the source link and sign up through Kaggle.