Deep Generative Networks

CS 585 is about getting deep networks to *produce* images and video rather than just classify them, learning the distribution of pixels well enough to synthesize, translate, inpaint, or extend them. Most of the weight sits on a semester-long project plus a presentation and report, with a midterm and final framing the theory behind GANs, transformers, and self-supervised reconstruction. It assumes you're comfortable with CNNs and basic deep learning from CS 464/484, and it's the natural follow-up if you want to do research in vision, multimodal models, or anything generative downstream.