Quantum computing

It has been theoretically proven that quantum computers can solve certain computational problems much faster than classical digital computers. In the future, quantum computers will also be able to tackle some previously unsolvable problems in artificial intelligence (AI) and machine learning (ML).

The simulation of complex systems and the solution of optimization problems are particularly promising areas of application. Simulations can be used, for example, to predict the properties of molecules and materials. Optimization problems arise, for example, in the energy industry when it comes to the optimal configuration of all available power plants (solar, wind, gas, etc.) to supply all consumers without overloading the grid and without producing excess energy.

At Fraunhofer IAIS, we have been developing new quantum algorithms for several years and researching the potential of quantum computing for practical applications. 

Our developments include the following areas:

• Resource-efficient quantum algorithms: The capabilities of today's quantum computers are subject to severe resource constraints. Naive formulations of quantum algorithms are not feasible on current and future generations of quantum computers due to the limited number of qubits, limited qubit connectivity, and limited circuit depth. We are researching resource-efficient alternatives, in particular decomposition techniques, to enable quantum algorithms to be executed for practical applications on today's quantum processors.
• AI-based synthesis of quantum circuits: Quantum circuits are programs for quantum computers. We are researching evolutionary and reinforcement learning-based methods for the automatic synthesis of quantum circuits. The focus is on optimizing Clifford circuits, as they are central subroutines of many algorithms and error correction in quantum computers.
• Explainability of quantum circuits: Many parametric quantum circuits follow a variation-based approach. Therefore, it is often unclear to what extent individual quantum operations are relevant for a specific application. We are investigating how explainable machine learning methods can contribute to the explainability of quantum circuits.
• State preparation and sampling: In order to load data into a quantum computer, a quantum state must be generated that represents the input data (state preparation). We are researching state preparation methods based on probabilistic graphical models, quadratic binary polynomials, and force-directed graph drawing.
• Experimental analysis of quantum computers: We test our algorithms on quantum processors of different technologies (superconducting, ion traps, neutral atoms, etc.) and manufacturers (IBM, IonQ, IQM, D-Wave, QuEra, etc.).
• Relation between classical and quantum machine learning methods: We investigate the relation between classical machine learning methods and corresponding quantum algorithms, including support vector machines, Markov random fields, restricted Boltzmann machines, and neural networks.

The research on quantum computing is closely linked to the Quantum Machine Intelligence Group at the Lamarr Institute. An important milestone is new insights into the expressiveness of quantum restricted Boltzmann machines.