Disentangled Diffusion Model

Disentangled diffusion models aim to separate underlying factors of variation within data, enabling controlled generation and manipulation of specific attributes. Current research focuses on applying this framework to diverse tasks, including dataset distillation, high-fidelity image and video generation (e.g., talking heads, textiles), and signal processing (e.g., ECG generation from PPG). This approach offers improvements in efficiency, controllability, and the quality of generated outputs across various domains, advancing both fundamental understanding of generative models and practical applications in areas like healthcare and creative design.