Beijing – – September 11, 2026 -- WiMi Hologram Cloud Inc. (NASDAQ: WiMi) has developed a multi-dimensional data pooling technique built on Variational Quantum Algorithms (VQA) that compresses complex datasets while retaining local feature detail, a persistent bottleneck in classical machine learning pipelines.
WiMi combines Quantum Haar Transform with partial measurement to eliminate flattening losses
The scheme merges a Quantum Haar Transform (QHT) with quantum partial measurement to build what the company calls a quantum pooling mechanism. Unlike classical pooling, which compresses data by discarding information through flattening, WiMi's method operates directly on multi-dimensional data in quantum state space, avoiding the need to reduce inputs to one-dimensional form before processing.
QHT maps high-dimensional data into quantum states using entanglement to preserve structure
QHT extends the classical Haar transform into a quantum computing framework, built on the universal quantum circuit used in quantum Fourier transforms. Each qubit in the system corresponds to one feature dimension of the input data, with quantum superposition coefficients encoding feature intensity. WiMi states this entanglement-based mapping preserves global structural information while reinforcing local feature correlations, addressing the exponential computational complexity that classical Haar transforms face when scaled to high-dimensional datasets.
Partial measurement replaces hard data discarding with probabilistic feature extraction
Instead of discarding data outright, WiMi's approach applies partial measurement to qubits using measurement bases matched to specific pooling strategies. For max-pooling, the measurement basis is designed to maximize the collapse probability tied to the strongest feature signal; for average-pooling, orthogonality constraints produce a probability-weighted average. Unmeasured qubits remain in superposition, maintaining continuity of local feature correlations, while measured outputs are converted into low-dimensional classical feature vectors.
Variational quantum algorithm optimizes circuit parameters to counter decoherence errors
A classical optimizer iteratively adjusts the parameters of a Parameterized Quantum Circuit (PQC) to minimize loss functions such as feature reconstruction error. WiMi says this optimization loop tunes both the QHT gate parameters and the measurement basis settings, aligning pooled outputs with downstream tasks including quantum classification and regression, while mitigating quantum decoherence errors during processing.
Company claims polynomial-level efficiency gains over classical high-dimensional pooling
WiMi asserts the technique delivers polynomial-level reductions in computational complexity compared with classical high-dimensional pooling algorithms, driven by quantum parallelism and QHT's orthogonality properties. The architecture is designed to adapt across data types by adjusting PQC gate structures, spanning one-dimensional audio, two-dimensional images, three-dimensional point clouds, and hyperspectral datasets.
WiMi positions the technology as groundwork for future quantum machine learning applications in computer vision, remote sensing, and biomedicine, contingent on continued advances in quantum hardware and algorithm refinement.