Fully Decentralized Semi-supervised Learning via Privacy-preserving Matrix Completion.

Area of Science:

• Machine Learning
• Distributed Systems
• Data Science

Background:

• Distributed learning involves inferring functions from data spread across nodes.
• Supervised and unsupervised distributed learning are well-studied, but distributed Semi-supervised learning remains less explored.
• Existing methods lack efficient and scalable solutions for distributed Semi-supervised learning.

Purpose of the Study:

• To propose a novel algorithm for distributed Semi-supervised learning.
• To extend the manifold regularization framework for this setting.
• To address the need for efficient, scalable, and privacy-preserving distributed learning.

Main Methods:

• Developed a fully distributed computation of the adjacency matrix for training patterns.
• Proposed a novel algorithm for low-rank distributed matrix completion using diffusion adaptation.
• Integrated flexible privacy-preserving mechanisms for similarity computation.

Main Results:

• The proposed distributed Semi-supervised algorithm is efficient and scalable.
• The method effectively computes the adjacency matrix in a distributed manner.
• Experimental results on standard benchmarks validate the algorithm's performance.

Conclusions:

• The novel distributed Semi-supervised learning algorithm effectively addresses the research gap.
• The approach offers an efficient, scalable, and privacy-preserving solution for distributed data analysis.
• The method demonstrates strong performance across various benchmarks, validating its utility.