Quantum Neural Networks Achieve Faster Gravitational Wave Data Analysis with 4 Qubits

Researchers are actively investigating whether cloud-based quantum computers can accelerate gravitational wave data analysis, a crucial task for the upcoming LISA space mission! Maria-Catalina Isfan, Laurentiu-Ioan Caramete, and Ana Caramete, all from the Institute of Space Science, INFLPR Subsidiary, University of Bucharest, have been evaluating the performance and cost-effectiveness of state-of-the-art quantum neural networks (QNNs) on platforms like IonQ and IQM Computers! Their work demonstrates that QNNs learn patterns in gravitational wave data significantly faster than classical networks, but initial cost analyses reveal substantial financial barriers , running even a small segment of the QNN can incur fees ranging from £2,000 to over £1,000,000! Despite challenges with device availability and software compatibility, the team achieved promising results, including 99% fidelity on a 3-qubit feature map and 20% prediction accuracy on a 4-qubit segment, highlighting the potential , and the hurdles , of utilising cloud quantum computing for real-time gravitational wave signal detection.

This represents a crucial step towards validating the potential of quantum machine learning for space-based gravitational wave astronomy. Despite these setbacks, the results, accumulated throughout 2024, provide valuable insights into the practical considerations of deploying QNNs on real quantum hardware. This work establishes a crucial baseline for evaluating the feasibility of quantum-enhanced data analysis pipelines for future space missions, offering a hardware-oriented approach to complement existing simulation-based studies. The QNN’s architecture features a 1st order Pauli expansion circuit for input encoding and a Pauli Two-Design circuit repeated four times, culminating in the measurement of all four qubits. The sQNN boasts 48 or 64 trainable parameters, depending on the number of qubits, while the lQNN consistently employs 64 parameters.

LISA QNN Deployment and Cloud Cost Analysis is

This study pioneers a hardware-oriented approach, moving beyond simulations to assess the viability of current quantum devices for LISA-relevant QNN architectures. Researchers meticulously analysed pricing structures from hardware providers including IonQ, IQM Computers, Amazon Braket, and Microsoft Azure, discovering significant cost variations, the first segment of their QNN would accumulate fees of £2000, £60000, and £1000000 respectively across these platforms. Input data was encoded using a 1st order Pauli expansion circuit, and the ansatz consisted of a repeated Pauli Two-Design circuit, ultimately measuring all four qubits at the circuit’s conclusion. This QNN is trained in two distinct steps, a short QNN (sQNN) and a long QNN (lQNN), each utilising different data pre-processing methods and dataset sizes. All results presented in this work were accumulated throughout 2024, providing a current snapshot of the landscape. This approach enables a realistic evaluation of quantum-enhanced data-analysis pipelines for future space missions, offering valuable insights into practical limitations and costs.

LISA QNN Deployment Costs on Quantum Hardware are

The research, conducted throughout 2024, focused on assessing the capabilities of current quantum hardware to support LISA-relevant QNN architectures and identifying practical limitations like runtime costs and noise levels. Experiments revealed the first segment of the QNN incurs costs of 2000, 60000, and 1000000 respectively when run on hardware from various providers. These results demonstrate the potential for quantum computation in accelerating LISA data analysis pipelines. The lQNN, employing 4 qubits, also features 124 gates and a similar circuit depth. Training the sQNN involved approximately 160 epochs with 5406 training samples and 1082 prediction samples, while the lQNN required around 170 epochs with 63062 training and prediction samples. Both QNNs utilise the Cobyla optimizer and possess 48 or 64 trainable parameters, depending on the number of qubits. Despite these setbacks, the team’s work provides valuable insights into the practical considerations of deploying QNNs on real quantum hardware for space-mission applications, paving the way for future quantum-enhanced data analysis pipelines.

Quantum LISA analysis costs and fidelity are significant

However, analysis of hardware provider costs, including Amazon Braket and Microsoft Azure, revealed substantial financial barriers, with fees for the initial QNN segment reaching up to £1,000,000! Despite applications to other providers like Pasqal and Munich Valley, responses were largely absent, and existing access was hampered by issues with software development kit versions and device unavailability, particularly with Microsoft Azure, throughout 2024. This study highlights the current limitations of cloud-based quantum computing for quantum machine learning (QML) workloads! While shallow circuit executions demonstrate high fidelity, deeper circuits and full model training are presently impractical due to rapid fidelity degradation and considerable expense.

System-level challenges, including software incompatibilities, version conflicts, and slow access procedures, further complicate usability! The authors acknowledge that the current quantum hardware is best suited for proof-of-concept and exploratory experiments, rather than operational deployment within the LISA data analysis pipeline. Future progress hinges on improvements in hardware reliability, pricing models tailored to QML, which often involves numerous short circuits and a large number of shots, and a more streamlined software ecosystem! Until these issues are addressed, classical simulators will remain the primary tool for meaningful QML development.