Machine Learning

OVD-SaaS : une architecture de microservices pour des applications industrielles d'intelligence artificielle

Publié le

Auteurs : Jose Armando Hernandez Gonzalez

This thesis addresses the problem of reproducibility in scientific research from the point of view of academic researchers, publishers and industry. From a review of the state of the art of the activity of these actors we provide a gap analysis which includes aspects such as evaluating and rewarding reproducibility, tracking and controlling research artifacts, and best practices in open science projects. We contribute by proposing solutions to these identified problem with a concrete methodology which includes the definition of new identifiers to conveniently join together authors with scientific publications and their source code as a whole, as well as any associated data. We shall call this methodology OVD-SaaS (Online Verifiable Datascience, Software as a Service). This research is complemented with a reference implementation as a proof of concept, and we discuss the difference between demos with a short life cycle with complete applications focused to industrial applications. We provide some illustrative use cases to this purpose. Finally, we analyze the viability of the OVD-SaaS taking into account the needs and requirements of academic researchers, publishers and industry.