Auto Regressive Diffusion

Autoregressive diffusion models combine the strengths of autoregressive modeling (capturing sequential dependencies) with the flexibility of diffusion models (generating diverse outputs). Current research focuses on extending these models to handle complex data structures like graphs and sequences of images, often employing permutation-invariant techniques to address order sensitivity. This approach improves generation quality and efficiency across various tasks, including text-to-motion, multi-image generation, and graph generation, demonstrating significant advancements in generative modeling for diverse data types. The resulting improvements in model performance and efficiency have broad implications for various fields, including computer vision, natural language processing, and drug discovery.