Deepquantum Achieves Closed-Loop Integration of Three Quantum Computing Paradigms

Quantum machine learning stands to revolutionise fields from drug discovery to materials science, yet developing and implementing these algorithms remains a significant challenge. Jun-Jie He, Ke-Ming Hu, and Yu-Ze Zhu, alongside colleagues including Guan-Ju Yan and Shu-Yi Liang from Shanghai Jiao Tong University, now present DeepQuantum, a new software platform designed to bridge this gap. This open-source framework, built on the popular PyTorch platform, allows researchers to efficiently design and run hybrid quantum-classical models on standard computing hardware, while also simulating photonic quantum computers. DeepQuantum uniquely integrates three distinct approaches to quantum computation, circuit-based, photonic circuit-based, and measurement-based, offering a versatile toolkit for both specialised and universal quantum algorithm design and paving the way for more accessible and powerful quantum machine learning applications.

Chinese National Quantum Science Funding Sources

This document details extensive research funding and support for a project focused on quantum science and technology. Key funding sources include the National Key R and D Program of China, the National Natural Science Foundation of China (NSFC), the Innovation Program for Quantum Science and Technology, and the Science and Technology Commission of Shanghai Municipality (STCSM). Additional support comes from the Startup Fund for Young Faculty at SJTU, the Frontier Technologies R and D Program of Jiangsu, and the Zhiyuan Future Scholar Program. Specific acknowledgements note additional support for X.

DeepQuantum Unifies AI and Quantum Computing Paradigms

Scientists have developed DeepQuantum, a new open-source software platform designed to bridge artificial intelligence and quantum computing. Built upon the widely adopted PyTorch platform, this framework enables efficient design and execution of both hybrid quantum-classical models and variational algorithms on standard CPUs and GPUs. DeepQuantum uniquely integrates three major paradigms of computing, quantum circuits, photonic quantum circuits, and measurement-based quantum computing, providing robust support for specialized and universal photonic algorithm design. The team achieved closed-loop integration of these paradigms, a first for any framework, and supports large-scale simulations through tensor network techniques and a distributed parallel computing architecture.

DeepQuantum’s architecture features three core classes, QubitCircuit, QumodeCircuit, and Pattern, allowing users to construct and simulate quantum computations with flexibility. The QumodeCircuit class includes Fock, Gaussian, and Bosonic backends, catering to diverse simulation needs in photonic quantum computing, while the Pattern class facilitates investigation of measurement-based quantum computation. Researchers demonstrated the framework’s capabilities through comprehensive benchmarks, achieving approximate simulation of circuits with over 100 qubits on a single laptop. DeepQuantum leverages PyTorch’s native communication protocol to efficiently utilize multi-node, multi-GPU computational power, boosting large-scale quantum simulations. The framework supports algorithm design and mapping for time-domain multiplexed photonic quantum circuits, and incorporates built-in optimizers to enable on-chip training of quantum machine learning models.

Hybrid Quantum-Classical Algorithms with DeepQuantum

DeepQuantum represents a significant advance in the field of quantum computing, delivering an open-source software platform built upon PyTorch that integrates artificial intelligence with both classical and quantum computational methods. The platform supports a range of quantum backends, including Fock, Gaussian, and Bosonic systems, and uniquely enables closed-loop integration of circuit, photonic circuit, and measurement-based computing paradigms. This capability facilitates the design and execution of hybrid quantum-classical algorithms, crucial for extracting practical value from current noisy intermediate-scale quantum devices. The researchers demonstrate DeepQuantum’s power through large-scale simulations, leveraging tensor network techniques and a distributed parallel computing architecture.

These simulations showcase the platform’s ability to model complex quantum systems and perform demanding computations, such as the quantum Fourier transform. The team acknowledges the ongoing challenges in scaling quantum computers and highlights the increasing role of artificial intelligence in overcoming hurdles related to quantum error correction and circuit compilation. Future work will likely focus on expanding the platform’s capabilities and applying it to a wider range of problems, further bridging the gap between theoretical quantum advantage and tangible real-world applications.