With a project closely related to MATH+ research in EF1-7 and EF1-11, the team Notorious F.U.B. wins the quantum machine learning hackathon QHack 2021.

In February 2021, more than 400 teams met up online to compete in the quantum machine learning hackathon QHack. The intense competition consisted of two rounds: first, the participating teams had to solve twelve prepared quantum machine learning programming challenges. In the second phase, the teams were asked to draft, plan, and execute their own quantum machine learning projects in only five days.

After completing phase one with a perfect score, the team Notorious F.U.B., that is Peter-Jan Derks (FU Berlin), Paul Fährmann (FU Berlin), Elies Gil-Fuster (FU Berlin), Tom Hubregtsen (FU Berlin), Johannes Jakob Meyer (FU Berlin, University of Copenhagen), and David Wierichs (University of Cologne), impressed the jury with their project Trainable Quantum Embedding Kernels with PennyLane.

The project investigates how training strategies for kernel methods known from “classical” machine learning can be applied to Quantum Embedding Kernels, evaluates their performance, and explores mechanisms to mitigate imperfections—like stable performance against finite sampling and hardware noise—in today’s quantum hardware. The team implemented a general-purpose software module into PennyLane (a software framework intended for differentiable quantum programming) and used it to run demonstration experiments on real quantum hardware. Furthermore, they created a tutorial for self-education on the subject.

The hackathon project Trainable Quantum Embedding Kernels with PennyLane is closely related to MATH+ projects EF1-7: Quantum Machine Learning and EF1-11: Quantum and Classical PAC Learning. It was the foundation for a recent preprint by the members of the team.