Markus Rambach, of the University of Queensland in Brisbane, and collaborators are exploring whether photonic quantum processors can enhance machine learning with limited data through Quantum Optical Reservoir Computing (QORC). The hybrid approach uses boson sampling to transform data before it reaches a classical machine learning model, positioning quantum computing as an additional computational layer rather than a replacement.
This research suggests a shift in focus from demonstrating quantum computational complexity to investigating whether quantum computation can contribute useful features to broader workflows. The study evaluated QORC on image-classification tasks, including handwritten digits and biomedical images.
Quantum Optical Reservoir Computing for Data Enrichment
The University of Queensland team achieved improved image classification even with limited training data by using a quantum process to create enhanced data representations. This approach, dubbed Quantum Optical Reservoir Computing (QORC), uses boson sampling on a photonic quantum processor to generate what researchers call a representation of the input data. Unlike traditional machine learning, the photonic processor in QORC remains fixed, acting as a transformation layer while only a classical classifier undergoes training to distinguish between classes.
This fixed transformation is key to the method’s efficiency in data-scarce environments. QORC builds upon the established machine learning technique of reservoir computing, where data flows through a complex, unchanging system before being processed by a simpler model. The researchers compressed and encoded classical data into a photonic quantum circuit, then directed multiple photons through an optical network to induce interference.
The resulting detection patterns form the quantum fingerprint, a richer representation of the original data intended to improve the performance of subsequent classical analysis. This focus on enriching data representation marks a shift in perspective, moving beyond the question of whether a quantum processor can solve classically intractable problems to whether it can enhance existing computational workflows. The team deliberately tested QORC’s performance under conditions mirroring real-world challenges, including imbalanced datasets and imperfections in photon sources.
Across image-classification tasks involving handwritten digits and biomedical images, the quantum reservoir consistently improved the performance of a linear classifier baseline. “Can that quantum computation contribute something useful to a broader computational workflow?” asked Markus Rambach, highlighting the team’s focus on practical integration rather than purely demonstrating quantum complexity.
The researchers’ work suggests that quantum computing may find an early niche not in replacing classical machine learning, but in augmenting it, particularly when labelled data is limited or expensive to acquire, Quandela says. The study points to a specific potential role for quantum computing in machine learning: not replacing classical models, but providing an additional computational layer that can enrich how data is represented.
The company, founded in 2017, builds photonic quantum computers based on single photons emitted from semiconductor light sources, and offers open-source software and training resources. This integration is facilitated by MerLin, Quandela’s quantum machine learning framework that adds photonic quantum layers to PyTorch and scikit-learn models, running on GPU simulators or Quandela’s hardware.
The DARPA selection of Quandela for Stage An of the Quantum Benchmarking Initiative, announced earlier in 2026, signals a broader interest in the potential of photonic quantum computing for utility-scale applications. The team’s findings demonstrate that a quantum processor can provide a valuable, fixed transformation layer, improving performance even when the classical classifier remains the primary component undergoing training.
Boson Sampling Transforms Data into Quantum Fingerprints
Boson sampling’s inherent complexity isn’t the primary aim of this work. Instead, researchers are evaluating whether the resulting photonic output distributions can serve as valuable features for machine learning algorithms. In this hybrid model, classical data undergoes compression and encoding into a photonic quantum circuit, initiating a cascade of photon interference within an optical network.
This interference pattern forms the quantum fingerprint, a transformed representation then fed into a classical machine-learning model for final classification. The system’s design deliberately avoids training the quantum processor itself, establishing it as a fixed transformation layer while the classical classifier adapts to distinguish between data classes.
The potential benefits of this approach are particularly relevant when labelled data is limited or unevenly distributed. The researchers hypothesize that the quantum fingerprint can enhance data representation in scenarios where classical methods struggle. The team’s work suggests that the value of quantum computing may lie in its ability to augment existing classical workflows, rather than entirely supplant them.
Launched in 2025, MerLin provides photonic quantum layers compatible with PyTorch and scikit-learn, enabling faster execution of quantum machine learning models on both GPU simulators and Quandela’s hardware, the firm reports. A collaboration with CMC Microsystems, announced in 2026, will provide access to Quandela’s photonic quantum computing platform via the Quantum Computing Sandbox.
Simultaneously, a partnership with Mekdam Holding Group aims to bring photonic quantum computing to markets in the Gulf region. These strategic alliances emphasise Quandela’s commitment to making quantum technology more accessible and integrated into broader computational infrastructures. This approach offers a potential pathway for using quantum computing in practical applications, even before fully fault-tolerant quantum computers become a reality.
QORC Improves Performance with Limited, Imbalanced Datasets
The photonic quantum reservoir computing approach yielded performance gains even with datasets mirroring real-world challenges, including limited training examples and skewed category representation. Researchers found the quantum component improved the ability of a classical classifier to accurately identify rare categories, an important advantage in applications like fraud detection and medical diagnostics where minority classes often hold the most significance. This was demonstrated using both artificially imbalanced datasets and naturally imbalanced biomedical imaging data, suggesting potential for practical application beyond controlled laboratory conditions.
The team’s work deliberately moved beyond ideal machine-learning benchmarks, testing the system with imperfect photon sources and limited data. This focus on realistic constraints distinguishes the research, as it addresses the practical hurdles of deploying quantum machine learning in real-world scenarios.
While a highly optimized deep neural network ultimately outperformed the hybrid quantum-classical system, the benefit of QORC was strongest when paired with simpler classical classifiers, indicating a potential role as a specialized component within existing workflows.
“The opportunity is not that quantum computing universally outperforms classical machine learning,” the researchers explain, “Instead, this research explores whether a photonic QPU can become a useful specialized component within a hybrid workflow, particularly under constraints such as limited or imbalanced training data.” Comparisons with classical feature-mapping techniques, such as Random Fourier Features, revealed similar accuracy improvements on the MNIST benchmark, helping to define the specific advantages of the quantum approach. The study does not propose replacing classical machine learning architectures, but rather augmenting them with a quantum layer that acts as a fixed transformation, leaving the training process to the classical component.
Ascella QPU Validates Hybrid Quantum-Classical Workflow
Quandela’s Ascella photonic quantum processing unit (QPU) demonstrated a performance advantage in image classification tasks when training data was limited, validating a hybrid quantum-classical workflow detailed by researchers at the University of Queensland. The team, led by Markus Rambach, observed improved accuracy on benchmark datasets using Quantum Optical Reservoir Computing (QORC), a method that uses photon interference to enrich data representations before they reach a classical machine learning model.
This approach specifically addresses challenges arising from scarce or unevenly distributed labelled data, a common bottleneck in many machine learning applications. The QORC workflow distinguishes itself by employing a fixed transformation layer within the quantum processor, rather than attempting to train the quantum system itself. Classical data undergoes initial compression, then is encoded into a photonic quantum circuit where photon interference generates a more complex feature set.
The researchers successfully ran this workflow on the Ascella QPU, confirming the benefits observed in prior numerical simulations and demonstrating the potential for practical implementation on physical quantum hardware. Quandela’s commitment to accessibility extends beyond hardware availability; the company offers the open-source MerLin framework, a quantum machine learning tool that integrates photonic quantum layers with PyTorch and scikit-learn models. This software, alongside Quandela’s free interactive training resources, lowers the barrier to entry for researchers and developers interested in exploring photonic quantum computing.
The initiative aims to define a concept for a utility-scale, fault-tolerant quantum computer, positioning Quandela as a leader in efforts to scale quantum computing beyond current limitations, the company states. The company, founded in 2017 and headquartered in Massy, France, has raised over €107 million, reflecting growing investor confidence in its photonic qubit technology and its potential to address real-world machine learning challenges.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
