3 problems

Convex duality conjecture. A similar convex-duality analysis can also be applied to deeper networks and other GANs.

Let $\mathcal{G}$ be a generator class, $\mathcal{F}$ a discriminator class, and let $\nu$ be a target density. Write $(\mathcal{G}_\star,\mathcal{F}_\star)$ for a generator/discri…

Consider the SWGAN methods and the latent representations described for learning handwritten digits, with the cost function applied either directly in pixel space or to a lower-dim…