DSS-GAN (Directional State Space GAN): A Generative Adversarial Network with Directional State Spaces Utilizing the Mamba Architecture

• Aleksander Ogonowski

Room: https://cern.zoom.us/j/66151941204?pwd=n7upvvZYibexBhbtyn5kvTpy36L0Wo.1 DSS-GAN is the first generative adversarial network to utilize the Mamba architecture as a hierarchical generator, synthesizing images from noise under class-conditional settings. The key contribution is the Directional Latent Routing (DLR) mechanism, which decomposes the latent vector and class signal into direction-specific sub-vectors, each independently conditioning the corresponding scanning direction along various spatial axes of the feature map.

Discussion Room: https://cern.zoom.us/j/66151941204?pwd=n7upvvZYibexBhbtyn5kvTpy36L0Wo.1