Privacy-Preserving Parallel Non-Negative Matrix Factorization with Edge Computing

Non-negative Matrix Factorization (NMF) is a computationally intensive matrix operation that resource-constrained clients struggle to complete locally. Privacy-preserving outsourcing allows clients to offload heavy computing tasks to powerful servers, effectively solving the problem of local computing difficulties. However, the existing privacy-preserving NMF outsourcing schemes only allow one server to perform outsourcing computation, resulting in low efficiency on the server side. In order to improve the efficiency of outsourcing computation, we propose a privacy-preserving parallel NMF outsourcing scheme with multiple edge servers. We adopt the matrix blocking technique to divide the computation task into multiple subtasks, and design the NMF parallel computation algorithm based on the multiplication updating rule. The proposed scheme implements the parallel outsourcing of non-negative matrix factorization based on multiple edge servers. We use random permutation matrices to encrypt original matrix, thereby protecting data privacy. In addition, we utilize the iterative nature of the NMF algorithm for result verification. Theoretical analysis and experimental results prove the advantages of the proposed scheme.