Researchers from UPenn and UIUC introduce Normalizing Trajectory Models (NTM), which use conditional normalizing flows to retain exact likelihood in few-step diffusion sampling. By combining shallow invertible blocks with a deep parallel predictor, NTM achieves state-of-the-art text-to-image results in just four steps while enabling self-distillation through its precise trajectory likelihood.
Normalizing Flow
归一化流
A generative model that maps simple distributions to complex target distributions via invertible transformations
Curated Intelligence
A new primer has been released that connects major generative AI architectures like PCA, VAEs, diffusion models, and GANs through compact mathematical derivations. Targeted at researchers and students with math backgrounds, the book focuses on revealing the unified mathematical structures behind these models rather than engineering implementation details.
FFJORD introduces a reversible generative architecture combining free-form continuous dynamics with flow models. The research aims to achieve exact likelihood estimation for high-dimensional data via Neural ODEs while maintaining computational efficiency and scalability.
OpenAI introduces Glow, a reversible generative model that simplifies previous flow-based architectures using invertible 1x1 convolutions. The model enables efficient high-resolution image generation and feature discovery for data manipulation, with code and visualization tools released.