Enhancement on the Labeled Component Unfolding System for Parallel Implementations.

Machine learning algorithms are in the fastestgrowing fields of interest in recent years. In this work, a semisupervised learning algorithm based on complex networks is adapted to exploit distributed processing of vertices. This is an algorithm based on the propagation of particles generated by labeled vertices to other edges of this complex network. Particles are absorbed at the edges and provide a dominance relationship between the vertices of the network and how the classes make up the problem. However, this algorithm is formulated by some sums that are performed throughout the database, which decreases the speed-up in parallel implementation due to underutilization. Our work focuses on stipulating an estimator of these values of this sum that decreases the need of the reduction. We demonstrate the formulation of the estimator and we show that the modification increases the classification performance of the original algorithm while increasing the parallelization potential of the algorithm.