Servindo modelos de machine learning com uma arquitetura baseada em serverless

Abstract

Smart solutions for data classification that make use of Deep Learning are in a moment of ascension. The data analysis area is attracting more and more developers and researchers, but the solutions developed need to be modularized into well-defined components in order to be able to parallelize some stages and obtain a good performance in the execution stage. From this motivation, this work presents a generic architecture for data classification, named Machine Learning in Serverless Architecture (MLSA), that can be reproduced in a production environment. In addition, the use of the architecture is presented in a project that makes multi-label classification of images to recommend tourist attractions and validates the use of serverless to serve models of Machine Learning. When using this type of approach, a decrease of at least 60% in processing time has been achieved compared to a monolithic approach.