Researchers from the group of theoretical physicist Hans Briegel have collaborated with NVIDIA to develop an AI method that automatically generates efficient quantum circuits, a key bottleneck in making existent quantum computers practically useful.
Before a quantum computer can perform any useful task, a quantum algorithm needs to be translated into a sequence of elementary quantum operations, known as quantum gates. Writing these quantum circuits efficiently is one of the hardest open problems in the field. Two years ago, a team led by Gorka Muñoz-Gil from the Department of Theoretical Physics unveiled a novel method to prepare quantum operations on a given quantum computer, using a machine learning generative model to find the appropriate sequence of quantum gates to execute a quantum operation.
Now the Innsbruck researchers, working with NVIDIA and the NVIDIA CUDA-Q platform for quantum-classical supercomputing, have refined this model to generate efficient quantum circuits. The method's key innovation is that it handles both the structural choices of a circuit (which gates to use) and their numerical parameters simultaneously. The resulting circuits are significantly shorter than those produced by competing methods, which is crucial for today's noisy quantum hardware where every additional gate introduces errors.
In a striking demonstration of the method's capabilities, the AI independently rediscovered the textbook circuit for the Quantum Fourier Transform, a fundamental building block of many quantum algorithms, without being told what the solution should look like. This demonstrates the promising use of AI across the quantum computing stack.
The work, recently published in Machine Learning: Science and Technology, was financially supported by the Austrian Science Fund FWF and the European Research Council.
If you would like to learn more, check out the CUDA-Q tutorial .
Publication: Synthesis of discrete-continuous quantum circuits with multimodal diffusion models. Florian Fürrutter, Zohim Chandani, Ikko Hamamura, Hans J Briegel, and Gorka Muñoz-Gil. Mach. Learn.: Sci. Technol. 7 025065 DOI: 10.1088/2632-2153/ae5b21
