WiMi Hologram Cloud Inc. (NASDAQ: WIMI) announced it is exploring a quantum machine learning algorithm designed for efficient training of large-scale machine learning models through quantum acceleration technology.
The algorithm uses classical machine learning methods to pre-train dense neural networks before constructing sparse neural networks to reduce computational requirements. WiMi developed a quantum ordinary differential equation system that requires sparsity and dissipation conditions to enable quantum acceleration.
The company employs quantum Kalman filtering to linearize nonlinear equations by transforming quantum state evolution into linear differential equations. This approach addresses quantum noise and other disturbances during computation.
After solving the quantum system, measurements are taken to obtain training parameters used to construct and optimize classical sparse neural networks. The quantum measurement process allows the quantum acceleration benefits to transfer to classical machine learning models.
The Beijing-based holographic augmented reality technology provider stated the algorithm combines sparsity with quantum acceleration to reduce computational complexity and improve training efficiency for large-scale models. WiMi indicated the approach could potentially lower energy consumption compared to traditional machine learning model training processes.
The company suggested potential applications include accelerated image and video processing for digital art and faster training of language models for natural language processing tasks. WiMi provides holographic cloud solutions including automotive HUD software, 3D holographic pulse LiDAR, and holographic semiconductor technology.