High Performance Computing for Auto Supervised Machine Learning Training: Parallel-Distributed Implementation of the Word2Vec Algorithm for Training Word Embeddings
摘要
This article presents a parallel-distributed implementation of the Word2Vec algorithm for training word embeddings, a relevant problem in natural language processing. A client-server implementation is developed, adapted to execute on non-dedicated commodity clusters. The server implements load balancing to deal with different processing times of processes and nodes. The experimental evaluation of the developed implementation was performed on a commodity cluster of workstations, analyzing the performance for different problem instances and varying the number of computing resources. Accurate speedup values and a proper scalability behavior are reported.