DeepQuantum: A PyTorch-based Software Platform for Quantum Machine Learning and Photonic Quantum Computing

Jun-Jie He ¶ \lx@paragraphsign, Ke-Ming Hu, Yu-Ze Zhu, Guan-Ju Yan, Shu-Yi Liang, Xiang Zhao, Ding Wang, Fei-Xiang Guo, Ze-Feng Lan, Xiao-Wen Shang, Zi-Ming Yin, Xin-Yang Jiang, Lin Yang, Hao Tang, Xian-Min Jin · 2025

Read DeepQuantum: A PyTorch-based Software Platform for Quantum Machine Learning and Photonic Quantum Computing by Jun-Jie He ¶ \lx@paragraphsign, Ke-Ming Hu, Yu-Ze Zhu, Guan-Ju Yan, Shu-Yi Liang, Xiang Zhao, Ding Wang, Fei-Xiang Guo, Ze-Feng Lan, Xiao-Wen Shang, Zi-Ming Yin, Xin-Yang Jiang, Lin Yang, Hao Tang, Xian-Min Jin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-18995.txt.

Introduction

Jun-Jie He and four co-authors introduce DeepQuantum, a free software platform built on PyTorch for quantum machine learning and light-based quantum computing. The platform runs hybrid quantum and classical models on normal processors and graphics chips. It also supports methods where measurement itself drives the computation.

For light-based work, three backends handle different kinds of photonic states and circuits. The authors say the platform is the first to link circuit, photonic, and measurement-based styles in one closed loop. Large jobs can use tensor network tricks and spread across many machines. Benchmarks and worked examples show what the platform can do in practice.

Catalog record

CREATORJun-Jie He ¶ \lx@paragraphsign, Ke-Ming Hu, Yu-Ze Zhu, Guan-Ju Yan, Shu-Yi Liang, Xiang Zhao, Ding Wang, Fei-Xiang Guo, Ze-Feng Lan, Xiao-Wen Shang, Zi-Ming Yin, Xin-Yang Jiang, Lin Yang, Hao Tang, Xian-Min Jin
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE95 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-18995

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

DeepQuantum: A PyTorch-based Software Platform for Quantum Machine Learning and Photonic Quantum Computing Jun-Jie He ¶ \lx@paragraphsign, Ke-Ming Hu, Yu-Ze Zhu, Guan-Ju Yan, Shu-Yi Liang, Xiang Zhao, Ding Wang, Fei-Xiang Guo, Ze-Feng Lan, Xiao-Wen Shang, Zi-Ming Yin, Xin-Yang Jiang, Lin Yang, Hao Tang, Xian-Min Jin Abstract We introduce DeepQuantum, an open-source, PyTorch-based software platform for quantum machine learning and photonic quantum computing. This AI-enhanced framework enables efficient design and execution of hybrid quantum-classical models and variational quantum algorithms on both CPUs and GPUs. For photonic quantum computing, DeepQuantum implements Fock, Gaussian, and Bosonic backends, catering to different simulation needs. Notably, it is the first framework to realize closed-loop integration of three paradigms of quantum computing, namely quantum circuits, photonic quantum circuits, and measurement-based quantum computing, thereby enabling robust support for both specialized and universal photonic quantum algorithm design. Furthermore, DeepQuantum supports large-scale simulations based on tensor network techniques and a distributed parallel computing architecture. We demonstrate these capabilities through comprehensive benchmarks and illustrative examples. With its unique features, DeepQuantum is intended to be a powerful platform for both AI for Quantum and Quantum for AI. $\lx@paragraphsign$ $\lx@paragraphsign$ footnotetext: [email protected] I Introduction Artificial intelligence (AI) and quantum computing (QC), both concepts conceived in the last century, are today at vastly different stages of maturity. While AI has become a transformative technology driving real-world value across countless domains, QC is in a nascent phase, primarily demonstrating its potential through carefully constructed proofs of quantum advantage. The pursuit of this advantage, where quantum devices perform tasks intractable for the most powerful classical supercomputers, has seen remarkable progress in recent years. Landmark experiments, such as quantum random sampling on superconducting and photonic platforms arute2019quantum ; wu2021strong ; zhong2020quantum , have convincingly demonstrated this capability for specific, contrived problems. However, a critical and pressing challenge for the field is to translate these theoretical demonstrations into practical, real-world applications with tangible value. The transition from demonstrating quantum advantage to harnessing quantum utility is of paramount importance. In the current noisy intermediate-scale quantum (NISQ) era preskill2018quantum , this translation effort relies heavily on the synergy between quantum processors and classical computers. Hybrid quantum-classical algorithms have emerged as the most promising paradigm PhysRevA.106.010101 , where a quantum device acts as a specialized co-processor whose parameters are optimized within a classical post-processing and feed-forward loop. Prominent examples that exemplify this approach include the variational quantum eigensolver (VQE) for quantum chemistry peruzzo2014variational , the quantum approximate optimization algorithm (QAOA) for combinatorial optimization farhi2014quantum , and a variety of methods in quantum machine learning (QML), such as the quantum reservoir computing and quantum extreme learning machine ghosh2019quantum ; mujal2021opportunities . These works demonstrate how quantum circuits can be utilized as powerful models within classical AI frameworks to achieve a “quantum-enhanced” effect dunjko2016quantum ; havlivcek2019supervised . As the community looks beyond the NISQ era, the ultimate goal is the realization of fault-tolerant quantum computing (FTQC), which promises to unlock the full potential of seminal algorithms like Shor’s shor1999polynomial and Grover’s grover1996fast . The journey from noisy, intermediate-scale devices to robust, large-scale fault-tolerant machines is fraught with immense challenges, particularly in areas including quantum error correction (QEC) and circuit compilation. Increasingly, AI is being recognized as an indispensable tool to overcome these hurdles. For instance, machine learning, especially deep reinforcement learning, has shown significant promise in designing more efficient QEC decoders andreasson2019quantum ; nautrup2019optimizing and optimizing the complex process of compiling high-level quantum algorithms into low-level hardware instructions fosel2021quantum ; pozzi2022using . As established, AI is crucial for advancing QC, from optimizing NISQ-era utilities to paving the way for FTQC. Conversely, the nascent field of QML provides compelling evidence for the reverse. Early studies suggest that quantum algorithms could enhance AI by offering faster training, improved model performance, or reduced data requirements for specific tasks acampora2025quantum . This creates a powerful virtuous cycle: AI accelerates the development of quantum hardware, which in turn promises to revolutionize AI itself. This reciprocal relationship underscores a future where the two fields are deeply intertwined. Given this profound and growing interdependency, the need for a programming framework that deeply and seamlessly integrates AI and QC is paramount. Several pioneering frameworks have emerged to address this, such as PennyLane bergholm2018pennylane and TensorFlow Quantum broughton2020tensorflow , which successfully treat quantum circuits as differentiable components within classical machine learning pipelines. While these tools have significantly advanced the field of QML, researchers may face practical challenges when attempting to couple quantum circuits with the ever-evolving landscape of state-of-the-art AI models. Furthermore, they predominantly focus on the standard qubit-based circuit model. However, to pave a viable path toward FTQC, it is crucial to consider alternative computational models and hardware platforms. Photonic quantum computing, in particular, stands out as a leading candidate due to its inherent advantages, such as low noise profiles, operational simplicity, and its role as a natural “flying qubit” for networking. This versatility has spurred recent breakthroughs, including the generation of integrated photonic Gottesman-Kitaev-Preskill (GKP) qubits larsen2025integrated and demonstrations of modular, manufacturable computing platforms aghaee2025scaling ; psiquantum2025manufacturable . Many of these scalable architectures are designed around the measurement-based quantum computing (MBQC) paradigm raussendorf2001one , which is naturally suited for fault-tolerant computation. Existing specialized frameworks, such as Strawberry Fields killoran2019strawberry and Piquasso kolarovszki2025piquasso for continuous-variable (CV) quantum computation, Perceval heurtel2023perceval for discrete-variable (DV) photonic systems, and Graphix sunami2022graphix for MBQC simulations, have provided invaluable tools for their respective domains, but a unified platform that bridges these models while maintaining a native AI integration is still lacking. To address these challenges and bridge the existing gaps, we introduce DeepQuantum, a novel programming framework engineered to foster the deep synergy between AI and QC for the NISQ era and beyond. This paper is organized as follows: Section II introduces the DeepQuantum framework, highlighting its global architecture and key features. Section III validates such advantages through benchmark results. Sections IV , V , and VI respectively focus on quantum computation with qubits, photonic qumodes, and MBQC patterns, covering the background, API overview, and application examples. Section VII demonstrates large-scale quantum simulation through tensor network techniques and distributed methods. In Section VIII , we summarize the whole paper and outline the future outlook. II DeepQuantum Overview Figure 1: The architecture of DeepQuantum and its ecosystem, including high-level quantum algorithms, toolchain, and low-level hardware. The DeepQuantum platform is designed to bridge AI and QC, providing an integrated environment that supports both AI-enhanced quantum algorithms and quantum-enhanced machine learning. Built with an emphasis on efficiency and usability, it offers a high-performance programming framework and an intuitive interface for constructing quantum circuits, allowing developers to prototype and explore advanced quantum applications rapidly. Implemented in Python and leveraging the widely adopted deep learning framework PyTorch paszke2019pytorch , DeepQuantum is fully open-source 1 1 1 https://github.com/TuringQ/deepquantum and can be easily installed via pip: ⬇ $ pip install deepquantum The overall architecture of DeepQuantum is illustrated in Fig. 1 . As an efficient programming framework tailored for quantum machine learning and photonic quantum computing, it achieves a closed-loop integration of three major paradigms: quantum circuits, photonic quantum circuits, and MBQC. At its core, the framework provides three principal classes for quantum computation: QubitCircuit , QumodeCircuit , and Pattern . The QubitCircuit

Read the complete text