MicroCloud Hologram Inc. advances the development of deployable quantum computing with the release of quantum recurrent neural network (QRNN) technology designed for sequential learning tasks. Through a novel approach centered around quantum recurrent blocks (QRBs), the company, trading as HOLO on NASDAQ, has addressed critical engineering bottlenecks that hinder the practical application of quantum recurrent models in current noisy intermediate-scale quantum (NISQ) devices. This new architecture systematically addresses the challenges of mapping core mechanisms such as recurrence and time dependence onto a quantum framework and has the potential to enable quantum machine learning for real-world data analysis. “Existing partial quantum recurrent models either rely too much on idealized assumptions about quantum operations or have difficulty adapting to current quantum hardware,” the HOLO researchers explained, highlighting the need for hardware-friendly solutions. The QRB is designed as a modular repeatable unit to minimize coherence time consumption.
Quantum Recurrent Block (QRB) Architecture for NISQ Devices
The new architecture promises to overcome significant hurdles in building practical quantum neural networks that can process sequential data. MicroCloud Hologram Inc. Quantum neural networks theoretically offer advantages over classical networks due to quantum phenomena such as superposition and entanglement, but translating these advantages into functional systems has proven difficult, especially for tasks involving sequences, such as natural language processing and time series analysis. The core of the innovation lies in the QRB itself, dubbed a “highly structured, parameter-controlled quantum subcircuit module” designed to characterize information updates in a sequence. Unlike traditional quantum neural networks that rely on expansive circuitry, QRB prioritizes hardware efficiency. HOLO researchers focused specifically on the limitations inherent in existing superconducting and ion trap platforms, including the number of two-qubit gates, connectivity, and noise. This approach minimizes the demands on qubit coherence, which is a key constraint for NISQ devices.
Beyond the internal structure of QRB, the overall network architecture adopts an “interleaved stacking” method. This differs from traditional deep learning’s layer-by-layer approach. Instead, QRNN reuses the same QRB structure in both time and feature dimensions. This reuse significantly reduces the number of required quantum gates and prevents the circuit depth from increasing with sequence length. This is an important consideration due to the limited coherence time of current quantum hardware. The model also utilizes a hybrid quantum-classical training framework where quantum circuits handle complex mapping and dynamic evolution of sequential data while leveraging classical computing for parameter optimization. “By measuring the quantum state and constructing a differentiable loss function, a classical optimizer can gradually update the variational parameters of the QRB, continuously improving the model’s performance in prediction or classification tasks,” say the HOLO researchers. Initial results show that QRNN outperforms traditional recurrent neural networks in prediction accuracy, especially in capturing subtle changes in time-series data.
Interleaved stacking network reduces circuit depth
The pursuit of practical quantum neural networks has long been hampered by the limitations of current quantum hardware. Although theoretical advantages exist, applying them to noisy intermediate-scale quantum (NISQ) devices remains a major challenge. Existing quantum neural network designs often require large, complex circuits that quickly suffer from qubit decoherence, or loss of quantum information, making reliable computation difficult. MicroCloud Hologram Inc. (HOLO) has addressed this problem with a new architecture centered around Quantum Recurrent Blocks (QRBs) and an innovative network design that significantly reduces circuit depth. Central to HOLO’s approach is a shift from holistic circuit construction to modularity. Unlike traditional quantum neural networks, QRBs are specifically designed to minimize the demands on qubit coherence. This careful consideration allows QRB to maintain expressiveness while avoiding unnecessary entanglement operations, which is a key element in mitigating the effects of decoherence. This reuse is especially important for NISQ devices where coherence time is the main limitation.
Compared to classical neural networks, quantum neural networks can exploit quantum superposition, entanglement, and high-dimensional Hilbert spaces to represent more complex functional structures under constrained parameter scales.
Hybrid quantum-classical variational optimization training
MicroCloud Hologram Inc. is actively working on the key challenge in quantum machine learning: turning theoretical possibilities into practical applications. HOLO researchers recognized that existing quantum recurrent models often become unstable when moving from simulation to real-world hardware due to stringent circuit depth and entanglement requirements. To overcome this, the team established three engineering principles to guide the creation of the QRB: modularity, reproducibility, and reduced coherence time. This is not an overall circuit, but rather a structured, parameter-controlled subcircuit designed to characterize the information updates at each step of the sequence. The physical implementation of each QRB utilizes a hardware-efficient gate set, carefully considering the limitations of both superconducting and ion trap quantum computing platforms. Rather than sequentially layering quantum recurrent blocks, HOLO reuses the same QRB in both the time and feature dimensions. “This design is particularly important for NISQ devices, as coherence time is typically the primary factor limiting the scale at which quantum algorithms can be implemented,” the company explains.
Unlike the global variational circuits commonly found in traditional quantum neural networks, QRBs are designed as highly structured, parameter-controlled quantum subcircuit modules to characterize the information update process at a single time step within a sequence.
Enhanced sequential learning performance and prediction accuracy
MicroCloud Hologram Inc. (HOLO) recently detailed its quantum recurrent neural network (QRNN), which is specifically designed to address the challenges of sequential learning, tasks that involve data with a unique temporal order, such as language processing and time series analysis. This is not just about applying quantum principles to existing architectures. This is a fundamental rethinking of how iteration, memory, and time dependencies are implemented on noisy intermediate-scale quantum (NISQ) devices. HOLO expects this technology to be the first to demonstrate the benefits of quantum in the near future and solidify the foundation for the industrialization of quantum artificial intelligence.
With the continued evolution of quantum computing hardware, this QRNN model is expected to be one of the first learning models to achieve quantum advantages in the near future, laying a solid foundation for the industrialization of quantum artificial intelligence.
