Quantum GANs for πππ―(ππ) Process Data Generation
Eraraya Ricardo Muten Togan Tlimakhov Yusuf Β· Zenodo (CERN European Organization for Nuclear Research) Β· 2021
In this report, we present the Deep Learning generative model GAN for the Higgs bosont ππΜ π―(ππΜ ) process data generation. Initially, a classical GAN model is considered, with Convolutional layers, Batch Normalization layers, and a Leaky ReLU activation function. The GAN aims to simulate the Higgs process precisely, capturing the crucial features in each b-jet produced. Two b-jets were considered in this work, each possessing four features that were resized to fit the Neural Network training process, where a relatively decent Wasserstein distance was obtained. Subsequently, a Quantum GAN model was considered, where the Quantum Circuit consisted of Gaussian gates as a continuous variable architecture per the nature of the dataset constraint. Xanadu's both PennyLane and Strawberry Fields Python libraries were used on a continuous variable quantum neural networkbased, where obtaining comparable results with the classical benchmark was intended on the simulators, considering a smaller dataset with fewer features.