Speaker
Description
Current research increasingly focuses on integrating parameterized quantum circuits into neural networks to create hybrid models capable of achieving performance competitive with classical counterparts—yet with fewer parameters—in machine learning tasks. However, optimizing the quantum architecture remains a bottleneck for the advancement of hybrid models. Our study proposes an efficient hybrid pipeline for speech enhancement, trained and validated in controlled and constrained environments. The results demonstrate the competitive performance of the proposed hybrid pipeline—comprising four parallel parameterized circuits and four qubits—which employs the Hybrid Cyclic Weight Sharing (HCWS) technique for optimal sharing of trainable quantum parameters across the quantum module, as well as the Data Re-Uploading (ReU) technique, which contributes to the model's expressiveness by inducing nonlinearity.
| Topic | Hybrid quantum-classical approaches for learning and optimization |
|---|---|
| Additional Topic | Quantum variational circuits and their applications |