Quantum Machine Learning

Quantum Machine Learning (QML) is an emerging discipline that makes use of the properties of quantum physics to solve challenges relating to machine learning and artificial intelligence. In this regard, QML offers some interesting advantages over conventional machine learning methods. One of the main advantages of quantum computers is their ability to perform a number of calculations simultaneously; this could exponentially improve the speed and efficiency of machine learning algorithms in the future. Quantum computers could also be better at recognizing patterns in complex datasets, including those that conventional computers are unable to distinguish from randomness. However, current hardware implementations of quantum computers have their technical limitations. For example, today's quantum computers only have about 50-100 qubits and can only execute algorithms of a limited length. In the long-term, it will only be possible to execute many of the developed algorithms using error-corrected hardware.

To make QML methods already applicable in the short and medium-term, the current focus of our research is on variational algorithms. These are quantum-classical hybrid methods in which parameterized quantum circuits are tailored to the particular application via a classic optimization loop. This allows a quantum computer to be configured for specific applications and to address the challenges of the problem at hand.

Our research focusses on identifying applications that can benefit from quantum computing. One area we are studying is the comparability of conventional and quantum-based algorithms. In addition, we are developing solutions to facilitate the use of these methods in automated and structured frameworks. Our work aims at pushing the boundaries of the state of the art in QML and finding new ways to use quantum computing in machine learning and artificial intelligence.

Our current focus in the field of QML involves the following topics:

  • Model Design: To process data with quantum computers, it must first be encoded into a quantum state. Aligning the encoding strategy with the specific problem is essential for the success of a QML pipeline. We are developing automated and systematic methods to create targeted algorithms.
  • Data: There are indications that the existence of benefits from quantum-based ML algorithms is determined not only by the algorithm itself but also particularly by the structure of the respective dataset. We are investigating for which data QML is especially suitable.
  • Automation: Both classical ML and QML require a high level of expertise. To simplify access to QML, we are working on the automated application of QML. This includes both the assessment of whether QML or classical ML is more appropriate for a given problem, as well as the specific selection of a suitable QML algorithm and the appropriate preprocessing.