© AI-generated
Quantum-generative models for industrial simulation systems
The development of quantum computers is progressing rapidly, but their use in industry is still in its infancy. The underlying technology has enormous potential to fundamentally improve industrial processes. As it becomes more powerful, it can be increasingly used in AI systems and significantly increases the precision and efficiency of industrial simulations.
Q-GeneSys translates findings from quantum supremacy experiments into practical solutions for industrial applications. The approach uses specific classes of quantum circuits, in particular IQP circuits, whose parameters can be trained classically, while the efficient sampling of high-dimensional probability distributions takes place on quantum hardware. In this way, complex tasks such as the generation of 3D designs, new molecular structures and product and system configurations are addressed where classical methods reach their limits. To ensure that quantum advantages can be used reliably and at an early stage, Q-GeneSys integrates fault tolerance right from the start - tailored to the circuit classes, the training methodology and the properties of current and future hardware.
Siemens AG, IQM Germany GmbH, OTH Regensburg and Fraunhofer IIS are working together iteratively: Industrial data sets and benchmarks define the fields of application, IQP-based model families are further developed and implemented in a hardware-aware manner, and methods for training, sampling and fault tolerance ensure practical usability. The aim is to design quantum-generative models in such a way that they are scalable, robust against noise and resource-efficient - with clear statements on the required hardware properties, overheads and the conditions under which a quantum advantage is realized.
The starting point is real industrial data sets from the fields of molecule, configuration and 3D design generation. Continuous and symbolic data are converted into binary representations, for example using Grammar Variational Autoencoder, so that their target distributions can be precisely compared with the model distributions of IQP circuits and classically optimized. The circuit parameters are adapted using well-defined target functions (such as KL divergence) with efficiently calculable expected values; quantum hardware then generates random samples from the learned distribution - conditioned if necessary to specifically ensure desired properties.
To maintain trainability and sampling complexity, Q-GeneSys continues to develop the circuit families: dynamic variants with intermediate measurements avoid "bar plateaus" without losing the complexity advantage. At the same time, early fault tolerance strategies are integrated: Where possible, classical methods limit error propagation, tailored quantum error correction protects critical operations with low overhead, partial error correction prioritizes the most sensitive circuit parts, and suitable error reduction further reduces residual errors. Realistic residual error models close the gap between ideal assumptions and real hardware so that training, decoding and sampling are robustly aligned. This co-design process of data, models, fault tolerance and evaluation leads to continuously refined, practical solutions - with clear resource and scaling analyses as a basis for decision-making.
Fraunhofer IIS (Quantum Compilation research group) leads the development of early fault tolerance and ensures that quantum generative models run reliably on current and future hardware - with minimal additional effort. We analyze the potential and limitations of classical error correction, design lean, IQP-specific quantum error correction including logical primitives and AI-supported decoding, and implement partial error correction that specifically protects the most critical circuit areas. In addition, we combine error correction with error reduction, model residual errors for realistic threshold and resource analyses and ensure the compatibility of dynamic circuits with stabilizer measurements. With proven expertise in quantum algorithms, AI-supported compilation and circuit knitting, we create the basis for realizing quantum advantages in 3D design, material and molecule development as well as product and system configuration in a practical and scalable manner.
Fraunhofer Institute for Integrated Circuits IIS