Гибридная архитектура с квантовой частью улучшила классификацию изображений на 50%

In July 2026, researchers presented on arXiv a hybrid system for image classification, combining quantum and classical components through a mixture of experts architecture. Based on testing results on MNIST and Fashion-MNIST datasets, the system reduced error rate by approximately half and demonstrated that quantum-inspired approaches are moving beyond theoretical research.

How the Hybrid System Works

The system is divided into two parts. The quantum part encodes the image through amplitude encoding, applies convolution operations through local unitary transformations, and processes data through several experts with different parameters. Feature extraction is performed using quantum stabiliser codes. Then the classical part combines the results of all experts through a fully connected neural network for final classification.

• Quantum part uses amplitude encoding to transform pixel data
• Architecture includes multiple experts, each processing the image with different parameters
• Classical part combines predictions from all experts into final prediction
• On MNIST and Fashion-MNIST, error reduction of ~50% was achieved
• GPU overhead remains acceptable for practical application on modern workstations

Results and Practicality

Joint analysis of experts showed better results than individual expert performance. Authors emphasize that the computational overhead of their quantum-inspired strategy is moderate on GPU workstations, making the approach a practical alternative to existing classical schemes. Additionally, researchers note that the quantum part of the framework can be executed on a real quantum processor — once such equipment becomes available.

What This Means

The research demonstrates that hybrid quantum-classical architectures with mixture of experts are beginning to transition from laboratories to practical applications. Thanks to moderate computational overhead, the approach already works on classical computers today, and with the development of quantum processors may show significant advantages.

Frequently Asked Questions

What is mixture of experts in this research?

Multiple experts — in this case quantum components — are trained on the same data with different parameters. Each expert processes the image independently, and then a classical neural network combines their predictions into the final result.

What datasets were used to test the system?

Researchers used MNIST (handwritten digits 28×28 pixels) and Fashion-MNIST (clothing images of the same size) — standard benchmark datasets for image classification tasks.

When can this be used on real quantum computers?

The architecture is already adapted to run on quantum processors, but this requires development of appropriate hardware. Currently, the system is practical for use on classical GPU workstations.