Researchers Generate States for Quantum Computing Via Boson Sampling

A new machine learning pipeline at Shahid Beheshti University, in collaboration with AriaQuanta Quantum Co and Shahid Sattari University of Aeronautical Sciences and Technology, accelerates the creation of Gottesman-Kitaev-Preskill (GKP) states, key resources for strong photonic quantum computing. Mohammad Amin Khanpour and Hossein Davoodi Yeganeh, alongside colleagues, present a two-stage surrogate model that accurately predicts the performance of Gaussian Boson Sampling circuits for GKP state generation, bypassing computationally expensive hafnian calculations. Achieving 90.0% GKP-detection accuracy and a 23.7 percentage-point improvement over existing methods, the approach sharply reduces the simulation burden by approximately 90%, representing a substantial step towards practical, all-photonic quantum computation.

Machine learning pipeline unlocks high-fidelity GKP states for scalable quantum computation

GKP-detection accuracy now reaches 90.0%, a 23.7 percentage-point leap beyond previous methods. This enables the creation of high-fidelity Gottesman-Kitaev-Preskill (GKP) states, essential for strong photonic quantum computing, which were previously unattainable due to computational limitations. Reaching this level of accuracy crosses a key threshold for error correction, as GKP states require a fidelity of at least 0.90 to meaningfully protect against logical errors in quantum calculations.

A new machine learning pipeline sharply reduces the computational burden of simulating Gaussian Boson Sampling (GBS) circuits, a technique for generating these non-Gaussian states, by approximately 90%. Previously, evaluating a single circuit configuration could take five minutes on a workstation. At 90.0%, accuracy in detecting Gottesman-Kitaev-Preskill (GKP) states represents a 23.7 percentage-point increase over previous techniques. These states are important for building stable photonic quantum computers, enabling more reliable encoding of quantum information and protection against errors.

The improvement was realised through a new machine learning pipeline that predicts optimal circuit configurations for Gaussian Boson Sampling (GBS), a method of generating these complex states using light. GBS utilises squeezed-state sources, linear interferometers and photon-number-resolving detectors. The pipeline reduces the computational time needed to assess a single circuit from five minutes on a standard computer to just milliseconds. Furthermore, the model’s predictions regarding circuit fidelity aligned with exact simulations, exhibiting a mean absolute error of only 0.032, and explained 83.7% of the variation in post-selection probability.

Accelerated Gaussian Boson Sampling via machine learning streamlines circuit design for

The pursuit of stable quantum computers hinges on creating and controlling qubits, the fundamental units of quantum information, and GKP states offer a particularly strong encoding scheme. Currently, however, this new machine learning pipeline operates effectively only with systems containing three to five optical modes. Scaling this surrogate model to the larger, more complex systems needed for practical quantum computation remains an open question.

It is important to acknowledge that this machine learning technique currently functions best with smaller quantum systems. A ninety percent reduction in simulation time represents a major step forward for Gaussian Boson Sampling, a method of building quantum computers using light. This accelerated evaluation allows designers to explore more complex circuits and refine designs more efficiently, despite current limitations. Ultimately, this work paves the way for more practical and scalable photonic quantum computation.

This new machine learning pipeline significantly accelerates the preparation of Gottesman-Kitaev-Preskill (GKP) states, a vital resource for building strong photonic quantum computers. GKP states encode quantum information in a way that is naturally resistant to errors. By accurately predicting the performance of Gaussian Boson Sampling (GBS) circuits, which generate these states using light and optical components, calculations previously required for design and validation were bypassed. Achieving ninety percent accuracy in GKP state detection, alongside a substantial reduction in simulation time, removes a key obstacle to scaling up photonic quantum processors.

The researchers developed a machine learning pipeline that accurately predicts the performance of Gaussian Boson Sampling circuits used to create Gottesman-Kitaev-Preskill states. These states are important because they offer a robust method for encoding quantum information, aiding the development of fault-tolerant quantum computers. The model achieved 90.0% accuracy in detecting these states across circuits utilising three to five optical modes, representing a significant improvement over previous methods. This accelerated evaluation of circuit designs allows for more efficient refinement and represents a step towards scalable photonic quantum computation.