Deep Neural Networks applied on sequence generation

• Grant number: 17/03706-2
• Support type: Scholarships in Brazil - Master
• Effective date (Start): June 01, 2017
• Effective date (End): May 31, 2019
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Cooperation agreement: Coordination of Improvement of Higher Education Personnel (CAPES)
• Principal Investigator: Eduardo Alves Do Valle Junior
• Grantee: George Gondim Ribeiro
• Home Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

In this project, we will apply Deep Neural Networks (DNNs) to Sequence Learning, a branch of Machine Learning that manipulates data chains - text, sound, video, and time series. The internet and, above all, social networks have caused an explosion of data, whose volume doubles each year, including sequential ones, creating both a challenge and a resource for the creation of automatic methods for its management. An important advance in these areas were DNNs, powerful models of machine learning that achieved excellent results in various Artificial Intelligence tasks. DNNs attract enormous attention in both the industry and the academy, with advances applied in a variety of practical fields. We will explore DNNs in Sequence Learning, trying to answer fundamental and applied questions with an empirical methodology. On the fundamental side, we will explore the limits of the generalization capacity of DNNs for the learning of sequences. On the applied side, we will evaluate the performance of DNNs in a recent proposed practical problem, the captioning of images. (AU)