Quantum machine learning provides solutions for industrial applications

Quantum computing is progressively evolving from a theoretical discipline into an applied one. Indeed, the development of a universal, programmable quantum computer has now moved beyond the realms of fiction and is gradually becoming a reality. This progress has huge implications for the solving of complex problems, as it paves the way for enormous improvements in performance – especially when it comes to optimization problems – thanks to the ability to pursue all computing pathways at the same time using entangled qubits, instead of one after another as in classical computers. At the same time, this is only possible if quantum mechanical phenomena such as the entanglement and superposition of microscopic states are used in a targeted way to extract information about the solution to a problem from the microscopic level and bring it into our macroscopic world, where it can be put to use. We harness the potential advantages of quantum technology to solve ubiquitous, industrially relevant decision problems and to optimize control tasks in manufacturing and industry.

Technology

Based on our expertise in the areas of machine learning, deep learning, and reinforcement learning, we develop hybrid technologies that use quantum computers in conjunction with conventional, classical computers. This serves to improve and accelerate the learning process for reinforcement learning, which is used to generate control strategies in dynamic and stochastic optimal control problems. In turn, this allows the identification of better approximative control strategies for time-critical and computationally intensive applications in a shorter time frame. 

Quantum computing-assisted reinforcement learning

As the implementation of quantum algorithms follows a completely different paradigm than normal programming approaches for conventional hardware, it follows that solutions based on quantum computing (QC) require different access technology. Research projects are developing both model-based and model-free QC-assisted reinforcement learning (RL) methods with a particular emphasis on the exploration of “noisy intermediate-scale quantum” (NISQ) technology and associated quantum algorithms. In addition, we exploit synergies between machine learning and variational – that is, adaptive or learning – quantum algorithms. This involves formulating industrially relevant optimal control strategies or problems as a Markov decision process (MDP) as part of modeling work. For example, the rapid and adaptive generation of approximate control strategies (policy) can be achieved using two QC-assisted RL algorithms: policy iteration and policy gradient methods. The components shown in blue correspond to a quantum algorithm and are executed on a quantum computer. Quantum machine learning (QML) methods or models are not only used to build the surrogate model but are also applied to policy approximation by variational quantum circuits (VQCs).

Fraunhofer IIS combines three areas of expertise that are essential for the development of hybrid systems: First among these is its expertise in key areas of machine learning, from classical methods to artificial intelligence (AI) and deep reinforcement learning. Secondly, it has expertise in the understanding of complex physical systems and processes, from the X-ray machine to the quantum computer. Lastly, we have also gained integral knowledge of industrial processes and requirements through countless industrial collaborations aimed at the profitable application of methods from the world of research.

Fraunhofer IIS helps you adapt machine learning processes to allow the efficient use of a quantum computer with a view to solving complex problems in your company processes more quickly. To facilitate calculations on quantum computers, we develop machine learning algorithms that reduce the error rates of the quantum computer, thereby delivering reliable results. As an R&D partner, we offer you not only our expertise in machine learning but also our knowledge of complex physical systems, as well as extensive industry knowledge gained through many years of practical experience.