SepRep-Net: Multi-source Free Domain Adaptation via Model Separation And Reparameterization
This addresses the problem of adapting multiple models to new domains without source data access for practitioners, offering an efficient solution that reduces computational overhead compared to ensemble methods.
The paper tackles multi-source free domain adaptation by proposing SepRep-Net, which reassembles multiple models into a unified network with separate pathways for training and reparameterizes them into a single one for inference, achieving competitive performance on target domains with low computational costs.
We consider multi-source free domain adaptation, the problem of adapting multiple existing models to a new domain without accessing the source data. Among existing approaches, methods based on model ensemble are effective in both the source and target domains, but incur significantly increased computational costs. Towards this dilemma, in this work, we propose a novel framework called SepRep-Net, which tackles multi-source free domain adaptation via model Separation and Reparameterization.Concretely, SepRep-Net reassembled multiple existing models to a unified network, while maintaining separate pathways (Separation). During training, separate pathways are optimized in parallel with the information exchange regularly performed via an additional feature merging unit. With our specific design, these pathways can be further reparameterized into a single one to facilitate inference (Reparameterization). SepRep-Net is characterized by 1) effectiveness: competitive performance on the target domain, 2) efficiency: low computational costs, and 3) generalizability: maintaining more source knowledge than existing solutions. As a general approach, SepRep-Net can be seamlessly plugged into various methods. Extensive experiments validate the performance of SepRep-Net on mainstream benchmarks.