The explosion of machine learning in the past two decades has already transformed large sectors of society, including healthcare, education, transport, and food and industrial production, as well as having an enormous impact on science and research. Indeed, the growth of deep learning, a type of machine learning, has been compared to the Cambrian Explosion of half a billion years ago when life on Earth experienced a short period of very rapid diversification.
Martin Jaggi, head of the Machine Learning and Optimization Laboratory, with his colleague Nicolas Flammarion teach the masters level Machine Learning Course, open to students from across campus. Recently, they have introduced two novel, practical elements to the course that have been welcomed by students and labs alike.
The first allows students to participate in the international “ML Reproducibility Challenge” a competition in which members of the machine learning community select a paper from a top ML conference and try to reproduce – and therefore validate – the experimental results described.
The second is the “Machine Learning 4 Science” (ML4Science) project component that is building cross campus collaborations, matching science projects from laboratories of all disciplines with students who will bring their machine learning expertise to new fields. Between 2018 and 2020 more than 600 students participated in projects proposed by 77 labs across EPFL, and even outside institutions including CERN.
“As the course is quite theoretical I really wanted it to be complemented by something a bit more practical. I think it’s fair to say that both students and labs feel that they benefit from the structure, it’s a real win-win,” says Jaggi. “It’s easy to lose track of the bigger picture when learning a new tool. Doing a real team project in an interdisciplinary setting, the students experience the diverse aspects that contribute to a project’s success – not just how many layers to put in a neural network.”
Professor Sahand Jamal Rahi, Head of the Laboratory of the Physics of Biological Systems in the School of Basic Sciences (SB) says the impact of ML4Science on his lab cannot be overstated. “I believe the experience is transformative for students who have an opportunity to break out of the classical classroom setting. They learn to work on highly challenging, cutting-edge problems instead of standard questions that do not change year-on-year, and face obstacles as they present themselves in research or industry, such as noisy and incomplete data or difficult-to-understand research articles in different fields. Martin’s students have figured out many of the ingredients that went into multiple papers and have changed how my lab does science,” he said.
Other work looked at an incredibly diverse set of research questions, including: predicting stroke severity using pacman game data; avalanche forecasting; music beyond major and minor; and, improving freshwater quality measurements.
More broadly, Professor John McKinney, Head of the Laboratory of Microbiology & Microtechnology in the School of Life Sciences (SV), believes the ML4Science approach is a brilliant example of a transition from passive to active learning and how EPFL should be restructuring traditional courses to better engage students.
“Not only does this course focus on acquisition and mastery of useful skills, it provides students with exciting opportunities to engage in cutting-edge scientific projects, allowing them to experience what it is actually like to be a scientist and to interact on a collaborative basis. Further, these projects can also provide the “glue” that brings together two or more labs with different areas of expertise, promoting the spirit and practice of interdisciplinary research. It provides a paradigm of how we should strive to restructure learning at EPFL more generally,” he said.
For Martin Jaggi the Award is a great honor. Both his parents are teachers and, although growing up he never imagined himself standing in front of a classroom full of students, the events of the past 18-months have made him realize what he is able to contribute.
“The beginning of the 2021/22 term has been very special because, after 18-months of distance learning due to COVID-19, I was looking out at 300 or so real students in the classroom again, with a great vibe of motivation and eagerness to learn. With ML4Science they get to take their first steps in a research project and learn how to use their tools in practice. There is so much talent at EPFL and I think it’s our responsibility to put this talent to use.”
The Credit Suisse Teaching Award is given each year to a person or an education team at EPFL, rewarding the best contribution to education within the Institution.