Diffusion Models

Forget Less by Learning from Parents through Hierarchical Relationships

Catastrophic forgetting is a persistent challenge for Custom Diffusion Models in continual learning, often addressed by minimizing interference between concepts. Instead, we exploit this very interference to learn common features among concepts.

Forget Less by Learning Together through Concept Consolidation

Catastrophic forgetting is a persistent challenge for Custom Diffusion Models in continual learning, often addressed by minimizing interference between concepts. Instead, we exploit this very interference to learn common features among concepts.

Learning Action Hierarchies via Hybrid Geometric Diffusion

Human actions unfold through multiple levels of abstraction, yet current action segmentation approaches typically consider a flat, Euclidean relationship between actions. This paper remedies this limitation by enforcing hierarchical structure amongst actions.