14.1 Quantum Generative Adversarial Networks
Generative Adversarial Networks (GANs) are powerful statistical techniques to generate (as much as needed) data close enough (in some sense) to given samples. They were introduced in [119] and originally tested on image data. Since then, they have seen wide applications in finance for time series generation [326, 327], tuning of trading models [190], portfolio management [211], synthetic data generation [17], and diverse types of fraud detection [285]. The gist of it is to have a generator and a discriminator compete against each other in order to improve themselves: the generator improves by becoming better at generating good samples (i.e., close to real data) from random noise, whereas the discriminator improves by being able to recognise real data from "fake" (namely generated) data. Both the generator and the discriminator are usually built as neural networks with hyperparameters over which...
