A Data Challenge for teaching Machine Learning, an educational innovation from Télécom Paris

At Télécom Paris, the teaching of Machine Learning includes a competition in which students solve a real business problem.

Machine Learning is a sub-category of Artificial Intelligence. It consists of letting algorithms discover recurring patterns in data sets. This data can be numbers, words, images or even statistics. “Machine Learning is such a vast field that classes are often either too theoretical, and not really in touch with real-life problems, or too practical, showing only a fraction of the existing methods,” laments Pavlo Mozharovskyi, research professor at Télécom Paris.

Together with Stéphan Clémençon, research professor in Statistical Learning, they established a 90-hour teaching unit that covers a wide range of this discipline with 14 theoretical classes supplemented by 14 practical classes, each dealing with a different method. But the most innovative part of the course is the ‘Data Challenge’, a competition in which students must solve a problem based on real company data.

This challenge takes place over a month, starting in the middle of the semester. “The company gives all the students the same set of data, along with a question to solve,” the teacher explains. Among the questions from previous years, we see, for example, “Check if the person in the two photos is the same” (IDEMIA company), or “Find an anomaly in the functioning of a helicopter engine” (Airbus).  Students must then develop an algorithm, which they test several times on the data set during the challenge. “They see their performance score immediately, and can thus gradually improve their code,” explains Pavlo Mozharovskyi. They must also submit a report explaining their development choices, with the 5 best works being presented and analyzed in a final session.

This course generally wins over the support of the participants, who give it scores of 4 and 5 (out of 5). “The competitive aspect encourages the students to put into practice what they have learned during the theoretical and practical classes, and to give the very best of themselves,” says Pavlo Mozharovskyi.

With more than 7 teachers and up to 10 PhD students, the Machine Learning course has been offered at Télécom Paris for six years and is continually updated to keep up with the new methods and tools of this large and constantly evolving field.

This is an educational innovation project submitted for the 2021 IMT Teaching Awards.

Illustrator: Lionel Tarchala