WiMi Hologram Cloud Inc. (NASDAQ: WiMi) has developed a Quantum Kernel Convolution (QKC) scheme designed to run on noisy intermediate-scale quantum (NISQ) devices, a surprising move from a company specializing in Hologram Augmented Reality technology. Rather than awaiting the arrival of fault-tolerant quantum computers, WiMi’s approach rethinks the computationally intensive process of convolution, specifically feature extraction and dimensionality reduction, by mapping image patches into quantum states and utilizing controlled entanglement evolution. WiMi points out that classical convolutional layers essentially rely on sliding windows and linear weighted summation to accomplish local feature extraction, whereas quantum computing inherently possesses the capability of high-dimensional Hilbert space representation and quantum parallelism. This hybrid Quantum Convolutional Neural Network (QCNN) integrates a quantum acceleration module for feature extraction within a classical deep learning framework, offering a practically feasible path toward quantum-enhanced image classification.
Quantum Kernel Convolution (QKC) for NISQ Device Implementation
WiMi Hologram Cloud Inc. This isn’t a distant vision of future quantum capabilities; the QKC scheme is specifically engineered to function on current noisy intermediate-scale quantum (NISQ) devices, sidestepping the need for the currently unavailable stability of fault-tolerant quantum computers. This focus on immediate implementation distinguishes WiMi’s approach from many other quantum machine learning initiatives. A key element of this approach is a novel pooling mechanism, which WiMi describes as an “information reallocation and selection mechanism that can achieve dimensionality compression without explicitly discarding information,” thereby lessening the computational load on subsequent quantum and classical circuits. At the architectural level, WiMi’s hybrid QCNN employs a layered design, strategically integrating quantum processing with established classical deep learning techniques. Classical neural networks handle preliminary data normalization, dimensionality adjustment, and final classification, while the quantum convolutional layer acts as a dedicated quantum acceleration module for feature extraction.
This synergy allows the model to benefit from mature classical toolchains while introducing quantum advantages at key computational nodes, avoiding the scalability limitations of fully quantum models given current hardware constraints. The implementation, built on the Qiskit framework, encapsulates the quantum convolutional layer as a reusable module, seamlessly integrating into existing deep learning workflows. WiMi details that during the training process, the model adopts a hybrid optimization strategy; classical backpropagation algorithms are used to update the parameters of the classical network, while the parameter-shift rule is utilized to estimate gradients for the quantum circuit parameters, achieving end-to-end joint training. Initial tests on the MNIST dataset demonstrate comparable classification accuracy to traditional CNNs, but with a significantly lower parameter count, suggesting a viable path toward practical quantum-enhanced image classification.
With a significantly lower number of parameters compared to traditional CNN models, this hybrid model is still able to achieve classification accuracy comparable to that of classical models.
