Personalized Generative Adversarial Nets(PGANs) for ECGIn this section, we introduce the framework of ECG GANoptimized for a specific patient ECG signal. We firstpresent a general framework for ECG signal generation usingGANs, adapted to the domain of ECG generation. Oneof the difficulties of creating realistic ECG is sustaining anatural medical cardiac morphology. We therefore devisea novel loss function for the task utilized by the generator.We discuss the details of implementation and optimizationof the generator and the discriminator. We call this adaptedGAN framework – ECG GAN. We then present the PersonalizedECG Generative Adversarial Network (PGAN) thatextends the ECG GAN with morphological signals derivedfrom patient-specific unlabeled data. The generated ECGsignals are then used to train a deep network (Section 5).We empirically show (Section 7) that the additional generatedlabeled examples significantly improve the ECG classificationin a patient-specific setting.