Detecting quantum complexity using transformer-based neural network (II)

Various data-driven attempts with classical agents have been made to analyze and capture the properties of quantum circuit, primarily to understand the region of quantum supremacy. Here, we also introduce a data-driven supervised-learning approach for our question of the quantum complexity of the given quantum circuit outputs. In particular, we focus on the transformer-based neural network model, consisting of several attention modules. With the conjecture that single circuit output will not solely demonstrate system property, a set of measured bit-strings from the wave function is fed into the neural network. As the experimental circuit data has a high level of noise, we first focus on the noiseless data to see whether there is a significant signal in the circuit data. We present a model performance on depth classification with varying noise levels, further discussing the possibility of quantum complexity.