[2602.00485] Replacing Parameters with Preferences: Federated Alignment of Heterogeneous Vision-Language Models

Computer Science > Artificial Intelligence arXiv:2602.00485 (cs) This paper has been withdrawn by Shule Lu [Submitted on 31 Jan 2026 (v1), last revised 5 Mar 2026 (this version, v2)] Title:Replacing Parameters with Preferences: Federated Alignment of Heterogeneous Vision-Language Models Authors:Shule Lu, Yujing Wang, Hainan Zhang, Xiaoshan Yang, Hongwei Zheng, Yongxin Tong, Changsheng Xu, Zhiming Zheng View a PDF of the paper titled Replacing Parameters with Preferences: Federated Alignment of Heterogeneous Vision-Language Models, by Shule Lu and 7 other authors No PDF available, click to view other formats Abstract:VLMs have broad potential in privacy-sensitive domains such as healthcare and finance, yet strict data-sharing constraints render centralized training infeasible. FL mitigates this issue by enabling decentralized training, but practical deployments face challenges due to client heterogeneity in computational resources, application requirements, and model architectures. We argue that while replacing data with model parameters characterizes the present of FL, replacing parameters with preferences represents a more scalable and privacy-preserving future. Motivated by this perspective, we propose MoR, a federated alignment framework based on GRPO with Mixture-of-Rewards for heterogeneous VLMs. MoR initializes a visual foundation model as a KL-regularized reference, while each client locally trains a reward model from local preference annotations, capturing specific...