WiMi Introduces RAE Quantum Feature Mapping, Improves MNIST Classification Accuracy
WiMi introduced Repeated Amplitude Encoding (RAE) to map classical data across qubit blocks, boosting expressive power while managing circuit depth and resources. In MNIST tests, quantum neural networks using RAE outperformed traditional amplitude and angle encoding in classification accuracy, convergence stability and robustness to parameter initialization.
1. Repeated Amplitude Encoding Technology
WiMi released the Repeated Amplitude Encoding (RAE) method, enabling repeated amplitude encoding of the same classical data across multiple qubit blocks. This new approach enhances the mapping capability of quantum neural networks while controlling circuit depth and qubit usage limitations.
2. Benchmark Results and Advantages
In experiments on the MNIST image classification dataset, quantum neural networks incorporating RAE outperformed models using traditional amplitude and angle encoding methods. The RAE-enabled networks demonstrated higher classification accuracy, improved convergence stability and greater robustness to parameter initialization under fixed resource conditions.
3. Implications for Quantum Neural Models
This technology addresses bottlenecks in current quantum feature mapping by leveraging repeated encoding to produce more discriminative representations. It lays an engineered path toward high-expressive-power quantum models that can be integrated into WiMi's quantum-enhanced holographic AR and computing solutions.