WiMi Hologram Cloud Inc. (NASDAQ: WiMi) is extending its expertise in holographic augmented reality into quantum computing with the release of Repeated Amplitude Encoding (RAE), a new technology designed to improve how classical data is processed by quantum neural networks. Existing quantum neural networks often struggle to fully utilize the potential of quantum states due to limitations in feature mapping, relying on mathematically linear transformations that restrict their ability to model complex data. WiMi’s RAE method addresses this by repeatedly encoding classical data across multiple qubit blocks, creating a pathway to higher expressive power while managing resource use. In experiments using the MNIST dataset, quantum neural networks incorporating RAE demonstrated improved classification accuracy, convergence stability, and robustness compared to traditional encoding methods, suggesting a significant step toward more capable quantum models.
Repeated Amplitude Encoding Enhances Quantum Neural Network Feature Mapping
WiMi Hologram Cloud Inc. This development addresses a critical limitation in current quantum machine learning models: the difficulty of effectively translating classical data into the quantum realm. Existing methods, reliant on linear or unitary transformations via parameterized quantum gates, struggle to fully leverage the exponentially high-dimensional space inherent in quantum states, hindering their ability to model complex relationships. The core innovation of RAE lies in its repeated encoding of classical data across multiple qubit blocks; this is not simply a refinement of existing techniques, but a fundamentally different method of quantum state representation. WiMi researchers recognized that traditional amplitude encoding, while efficient in qubit usage, often dilutes feature distributions during quantum circuit evolution, limiting the ability to model nonlinear structures. The company stated that they addressed these limitations by re-examining how classical data enters the quantum system, starting from the fundamental mechanism of quantum state representation.
To demonstrate RAE’s efficacy, WiMi employed the MNIST dataset, a standard benchmark for image classification, and integrated the method into several quantum neural network architectures. Results revealed a significant performance boost compared to traditional amplitude and angle encoding methods. Under fixed class conditions, networks utilizing RAE exhibited improved classification accuracy, convergence stability, and resilience to parameter initialization. WiMi reported that experimental results show that, under the condition of a fixed number of classes, quantum neural networks adopting repeated amplitude encoding outperform the control methods, suggesting that RAE enables more discriminative feature representations even with comparable task complexity. This advancement positions WiMi as a key player in bridging the gap between classical data processing and the potential of quantum machine learning.
MNIST Dataset Validates RAE’s Improved Accuracy and Stability
WiMi Hologram Cloud Inc. The core issue, as WiMi researchers identified, lies in the difficulty of effectively translating classical information into a format quantum computers can utilize. The company stated in a recent release that this technology effectively enhances the mapping capability of quantum neural networks to complex feature spaces, increasing expressive power alongside controllable resource usage. Researchers integrated the method into various quantum neural network architectures, comparing its performance against traditional amplitude and angle encoding techniques.
Although the quantum state itself resides in an exponentially high-dimensional space, in practical models, the limited encoding methods make it difficult to fully unleash this high-dimensional advantage, resulting in issues such as insufficient mapping capability and weak category scalability in complex classification tasks.
