WiMi Achieves Coexistence of Lightweight Design and High Performance by Efficiently Embedding Quantum Modules into U-Net

Core Insights - WiMi Hologram Cloud Inc. has developed a hybrid quantum-classical deep learning technology called QB-Net, which integrates lightweight quantum computing modules into the classical U-Net architecture, achieving a reduction in parameters by up to 30 times while maintaining comparable performance [1][9][10] Technology Overview - The core advantage of quantum computing is its ability to express high-dimensional information and perform operations in exponentially dimensional spaces, surpassing classical architectures [2] - WiMi's approach focuses on constructing quantum enhancement modules rather than fully quantized AI models, addressing the bottleneck layer of deep networks with quantum states that can express high-dimensional features efficiently [3][4] - QB-Net retains the U-Net structure but replaces traditional convolutional layers at the bottleneck with a quantum feature compression-transformation-reconstruction module, which consists of three key steps: encoding classical features into quantum states, transforming features through quantum circuits, and decoding quantum states back into classical tensors [5][8] Quantum Feature Transformation - The encoding step minimizes qubit usage while preserving key information, using techniques like linear projection or amplitude encoding [5] - The transformation step utilizes parameterized quantum circuits (PQC) to achieve expressive transformations with significantly fewer parameters compared to traditional convolutional layers [6][7] - The decoding step reconstructs the quantum measurement results into classical tensors, allowing for efficient filtering and abstraction of high-dimensional information [8] Strategic Implications - The introduction of QB-Net signifies a significant advancement in quantum AI technology, demonstrating the real value of quantum computing and its potential integration with deep learning [9] - The hybrid quantum-classical architecture is expected to become a mainstream form of AI, providing a new optimization paradigm for the global AI industry and enhancing enterprise-level intelligent systems [10]