Speaker
Description
Physics-Informed Neural Networks (PINNs) revolutionized the field of Scientific Machine Learning. The community has explored several variants of PINN. Shah et al. introduced Quantum-Assisted Physics-Informed Neural Network (QAPINN)$-$ a hybrid quantum-classical variant of PINN. QAPINN reduced weights up to 20\% as compared to the equivalent classical PINN while improving or maintaining accuracy. In this work, we employ activation maps as an initial explainability tool to investigate how QAPINNs learn and discuss future directions using richer representation analysis techniques. We discuss the effect of introducing a quantum hidden layer on the learning process of a PINN, the role of entanglement and output extraction method, and effect of choice of rotational gates for the variational layers in QAPINN architecture.
| Topic | Hybrid quantum-classical approaches for learning and optimization |
|---|---|
| Additional Topic | Quantum variational circuits and their applications |