A marvelous weekend of team building, learning and fun. That could be Kapernikov’s conclusion after participating at ArcelorMittal Belgium’s first hackathon in Ghent from 15 to 17 September 2017. The hackathon was a great way for Kapernikov to test the algorithms the team had been working on and to find a creative solution to a real-life problem that is typical for the steel industry.
The challenge for Kapernikov was to see whether its AI algorithms could do better than a human operator in predicting when a weld would break. This welding process is an important step in the so-called cold rolling process, which uses a high-precision mill to reduce the thickness of steel to customer specifications.
Kapernikov ran a Fast Fourier transformation on the raw data and reduced dimensions by means of PCA compression. Our final solution, based on SVM and a deep learning auto-encoder, ranked second-best of all teams in the competition category Artificial Intelligence. Unfortunately (although not surprisingly), within the short timespan of 36 hours, none of the teams actually did better than an experienced human operator would do.
Learning experience
“Above all, this was a fun experience, and a great opportunity to work with colleagues in a completely different setting,” says Ezechiël Syx, data consultant at Kapernikov. “For me personally, this was a great learning experience as well, and I picked up some great new tips and tricks which I can apply in my daily work at Kapernikov.”
The 65 participants of the hackathon were given 36 hours to come up with a creative solution for the proposed welding fractures problem. After these 36 hours, the award ceremony took place. Each team had to give a five-minute presentation to pitch their project to the panel of judges, consisting of experts from ArcelorMittal Belgium, Ghent University and the University of Liège.
Competition and more
Other side-activities included a workshop about pitching and one about TensorFlow™, an open source software library for numerical computation using data flow graphs. But the organization also paid attention to the fun factor, by including great food and drinks, a drone race, and the opportunity to network with friends and colleagues.
“The event was well organized, and I think we learned some lessons for next editions,” says Ezechiël Syx. “For one thing, having the algorithm run smoothly on a Microsoft Azure platform was a challenge for us. And to get the required results from certain AI algorithms, you probably need more than the 36 hours that were foreseen. But other than that, we are definitely eager to participate again in next editions of the ArcelorMittal Belgium hackathon.”