We study to find the link between the classical signal processing theorems and deep learning approaches. The analysis of the classical optimization problem proposed from the field of signal processing can be utilized as a novel loss function to train the deep neural network, which provides significantly improved performance. Signal processing-based learning algorithms can be a powerful model for image processing.

   In the real world, it is difficult to gather a lot of valuable data for training. Generative models such as generative adversarial network (GAN), and diffusion model can be a powerful breakthrough to address this issue. Not only for data generation but also for image-to-image translation and various vision tasks, we study to propose stable and robust generative models.