Layered Quantum Neural Networks Enhance Squeezing and Enable Faster Metrology in Field Sensors

Squeezing, a technique that reduces uncertainty in one physical property at the expense of another, significantly enhances the precision of measurements, and researchers continually seek ways to generate and manipulate this resource more effectively. Nickholas Gutierrez, Rodrigo Araiza Bravo, and Susanne Yelin investigate how the architecture of quantum neural networks can accelerate squeezing and improve the sensitivity of quantum sensors. Their work compares three quantum models, reservoir computers, perceptrons, and multi-layer neural networks, revealing that stacking perceptrons into a multi-layer network dramatically reduces the time required to achieve optimal squeezing. Importantly, this approach maintains Heisenberg-limited sensitivity, allowing for increasingly precise measurements, and the team demonstrates a scaling effect where adding more layers further enhances performance, representing a significant advance in the design of squeezing-based sensors.

The team investigates how different quantum architectures, including layered quantum neural networks, perform in sensing applications, utilising principles from quantum computation to improve measurement precision beyond classical limits. By carefully designing network connectivity and applying tailored quantum operations, scientists demonstrate enhanced squeezing levels and, consequently, faster measurement rates compared to traditional methods. The findings reveal that specific network configurations optimise the balance between squeezing and decoherence, leading to significant improvements in the signal-to-noise ratio for sensing applications. The researchers compared three architectures, quantum reservoir computers, quantum perceptrons, and multi-layer quantum neural networks, when used as squeezing-based field sensors. They employed a standard metrological sequence involving coherent-spin preparation, one-axis-twisting dynamics, field encoding via a weak rotation, time-reversal, and collective readout. The research demonstrates that a single quantum perceptron achieves the same optimal sensitivity as a quantum reservoir computer, but benefits from accelerated squeezing due to steering by the output qubit. Stacking perceptrons into a quantum neural network further amplifies this effect, and in a two-layer network with input and output qubits, optimal sensitivity is achieved.

Quantum Neural Networks Boost Metrology Performance This research explores how the architecture of a quantum system, specifically using structures inspired by quantum neural networks, can significantly enhance the performance of squeezing-based quantum metrology. It moves beyond simply focusing on squeezing strength and particle number, arguing that how qubits are connected and orchestrated in time is crucial for maximising sensing capabilities. Key findings demonstrate that quantum neural networks, built from interconnected perceptron-like layers, outperform traditional and basic layered approaches, such as quantum reservoir computers which squeeze all qubits simultaneously. The improvement stems from the sequential squeezing of layers, allowing for additive contributions to the metrological signal, and benefits scale with the number of layers, resulting in a root-L enhancement in sensitivity. Compared to a quantum reservoir computer, and a single perceptron, quantum neural networks offer superior performance. The research has implications beyond Hamiltonian design, suggesting that focusing on qubit arrangement and temporal orchestration is as important as designing the Hamiltonian itself. Leveraging machine learning-inspired structures, architectures developed for quantum computation can be repurposed for enhanced sensing, and the proposed approach is potentially implementable on current quantum hardware platforms, including neutral atoms, trapped ions, and superconducting qubits.

Neural Networks Boost Quantum Sensor Performance This research demonstrates that the structure of multi-layer neural networks can significantly enhance the performance of quantum sensors based on spin squeezing.

Scientists have shown that these networks, composed of interconnected qubits, achieve improved sensitivity and faster operation compared to traditional reservoir computer architectures. The key lies in the sequential application of squeezing, where each layer of the network both receives and imparts squeezed states to adjacent layers, resulting in an additive effect on the overall sensing capability. Specifically, a two-layer network exhibits a root-two, or approximately 1. 41, improvement in sensitivity over a comparable reservoir computer, while maintaining the fundamental limit of Heisenberg scaling. This advantage extends to networks with multiple layers, where the sensitivity increases with the square root of the number of layers, and the time required for optimal squeezing decreases as the number of qubits per layer increases.

The team found that the scaling of sensitivity for a multi-layer network is proportional to the square root of the number of layers, while a reservoir computer scales with the number of layers raised to the power of three-halves. The authors acknowledge that the performance gains are most pronounced when the number of qubits per layer is significantly larger than the number of layers, and future work will likely focus on optimising the network architecture and exploring the potential for even greater enhancements in sensitivity and speed, potentially paving the way for more precise quantum sensors for a range of applications. 👉 More information 🗞 Enhanced Squeezing and Faster Metrology from Layered Quantum Neural Networks 🧠 ArXiv: https://arxiv.org/abs/2512.09137 Tags: