Distributed Semisupervised Fuzzy Regression With Interpolation Consistency Regularization

Recently, distributed semi-supervised learning (DSSL) algorithms have shown their effectiveness in leveraging unlabeled samples over interconnected networks, where agents cannot share original data with each other and can only communicate non-sensitive information neighbors. However, existing DSSL cope uncertainties may suffer from high computation communication overhead problems. To handle these issues, we propose a fuzzy regression (DSFR) model if-then rules interpolation consistency regularization (ICR). The ICR, which was proposed recently for problem, force decision boundaries to pass through sparse areas, thus increasing robustness. its application scenarios has not been considered yet. In this work, Fuzzy C-means (DFCM) method (DICR) built on the well-known alternating direction of multipliers respectively locate parameters antecedent consequent components DSFR. Notably, DSFR converges very fast since it does involve back-propagation procedure is scalable large-scale datasets benefiting utilization DFCM DICR. Experiments results both artificial real-world show that achieve much better performance than state-of-the-art algorithm terms loss value computational cost.