Avoiding barren plateaus with classical deep neural networks

Variational quantum algorithms (VQAs) are among the most promising in era of Noisy Intermediate Scale Quantum Devices. Such constructed using a parameterization U($\pmb{\theta}$) with classical optimizer that updates parameters $\pmb{\theta}$ order to minimize cost function $C$. For this task, general gradient descent method, or one its variants, is used. This method where circuit updated iteratively gradient. However, several works literature have shown suffers from phenomenon known as Barren Plateaus (BP). In work, we propose new mitigate BPs. general, used $U$ randomly generated. our they obtained neural network (CNN). We show besides being able BPs during startup, also effect VQA training. addition, how behaves for different CNN architectures.