MicroCloud Hologram Inc. (NASDAQ: HOLO) shares climbed 4.51% to $1.6199 during Thursday morning trading after the company announced the development of a Deep Spiking Quantum Neural Network, dubbed DSQ-Net, designed specifically for classifying noisy images. The technology represents a deliberate step toward integrating quantum computing with neuromorphic engineering — two fields that have largely developed independently, with major players such as IBM, Google, and Intel pursuing separate quantum and brain-inspired computing research programs.

DSQ-Net Architecture

MicroCloud Hologram developed DSQ-Net building on years of accumulated quantum algorithm research and engineering. The system introduces a variational quantum circuit as a core auxiliary training mechanism — a notable departure from earlier approaches that leveraged quantum circuits solely for feature mapping.

The architecture embeds quantum circuits directly into the training processes of spiking neural networks. Spiking neural networks, which process information using discrete temporal pulses modeled loosely on biological neurons, have historically been difficult to train using standard gradient-based methods because their discrete firing events are non-differentiable. According to the company's engineers, this design confronts non-differentiable spiking events and random neuronal dynamics directly, restoring trainability for spiking networks at the system level.

Input images first pass through a classical preprocessing module that converts pixel data into spatio-temporal spike sequences. Within this framework, noise becomes part of the temporal signal rather than functioning purely as interference.

Quantum Layer and Accuracy Performance

The deep spiking network extracts high-level spatio-temporal features from encoded data. Intermediate spike statistics are mapped into qubit states through a technique known as amplitude encoding, which allows complex spike structures to be represented efficiently using fewer qubits.

A parameterized variational quantum circuit then further evolves the encoded quantum state. Multiple tunable quantum gates are updated in conjunction with the classical optimizer throughout the training cycle, with quantum measurement results guiding adjustments to classical network weights.

In testing, DSQ-Net maintained classification accuracy above 90% on previously unseen noisy images. The model outperformed classical spiking networks of comparable scale in direct comparisons, and performance degraded more gradually as noise levels increased, the company reported. These results were self-reported and have not been independently verified by third-party researchers.

Validation and Real-World Applications

MicroCloud Hologram validated the model using a high-fidelity quantum simulator rather than physical quantum hardware, a decision intended to ensure reproducible results while quantum hardware technology remains in a relatively immature stage. This approach is common across the quantum machine learning research community, where hardware limitations including qubit count, coherence time, and error rates still constrain what can be executed on physical processors. Researchers constructed multiple datasets featuring varied noise types and intensities to conduct thorough testing.

The company sees potential applications spanning industrial inspection, traffic perception systems, and constrained edge computing devices. Spiking networks are inherently suited for such use cases due to their low-power, event-driven processing characteristics. Quantum parallelism could further reduce overall computational complexity as quantum hardware matures — though moving from simulation to physical deployment would require hardware capable of reliably running the variational circuits at scale.

The breakthrough also pushes quantum machine learning beyond purely theoretical speed advantages, addressing practical challenges such as noise, uncertainty, and non-differentiable dynamics. MicroCloud Hologram positions DSQ-Net as a bridge connecting the quantum computing and neuromorphic computing fields. Whether the architecture can transition from simulator-based validation to results on near-term quantum hardware remains an open question for the broader field.

Source: Blockonomi