WiMi Unveils Quantum Deep Convolutional Neural Network for Image Recognition
WIMI•WiMi has developed a quantum deep convolutional neural network model integrating quantum parameterized circuits for image recognition, using a quantum-classical hybrid training scheme to boost computational efficiency. The architecture comprises data encoding, stacked quantum convolutional and feature fusion layers, and a quantum classification module leveraging parallel gate operations.
1. Quantum Deep CNN Breakthrough
WiMi has introduced a quantum deep convolutional neural network model that mirrors classical CNN hierarchies but replaces convolution kernels with parameterized quantum circuits. The design leverages rotation, control and entanglement gates to perform parallel feature extraction across superposed qubit states, aiming for higher expressive power in image recognition.
2. Data Encoding and Feature Fusion Modules
The system converts classical image pixels into qubit probability amplitudes via amplitude, angle or hybrid encoding strategies. After stacked quantum convolutional layers extract local features, a quantum feature fusion module uses entanglement gates to integrate information across qubits, enabling higher-dimensional representations with reduced overhead compared to matrix-based fusion.
3. Quantum-Classical Hybrid Training Scheme
WiMi’s hybrid training strategy executes forward computations on quantum circuits and uses classical optimizers for parameter updates. Measurement of key qubits yields error metrics, which classical algorithms process to adjust trainable parameters in the quantum circuit, addressing current hardware limitations while harnessing quantum parallelism.
