Analyzing the barren plateau phenomenon in quantum neural network training - Chen Zhao

Abstract:
In recent years, hybrid quantum-classical algorithms are widely used in quantum chemistry, combinatorial optimization, and quantum machine learning. These algorithms are regarded as practical quantum algorithms with potential quantum advantages in the NISQ era. In these algorithms, we usually need to train parameterized quantum circuits (PQCs), which are also often called quantum neural networks (QNNs). There are still many difficulties when training QNNs. One of the major difficulties is the barren plateau (BP) phenomenon, where gradients vanish exponentially in the system size. In this talk, we will introduce how to analyze whether there exist barren plateaus with a given PQC with the ZX-calculus. And we will use these techniques to analyze 4 different structures of PQCs, the hardware-efficient ansatz, the QCNN, the tree tensor network ansatz, and the MPS-inspired ansatz.

Recorded on January 25th 2021