Enneng Yang

This paper proposes a data-free dual orthogonal projection framework to perform continual model merging.

We propose a training-free projection-based continual merging method that efficiently combines models incrementally.