A study of image representations from multiple domains using unsupervised and semi-supervised deep learning

• Grant number: 19/02033-0
• Support type: Scholarships in Brazil - Master
• Effective date (Start): September 01, 2019
• Effective date (End): March 31, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal researcher: Moacir Antonelli Ponti
• Grantee: Gabriel Biscaro Cavallari
• Home Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

Deep neural networks for image processing and representation learning are currently applied with great success in tasks for which there is annotation available. Although some architectures are capable of generalizing for different problems, there is a gap in the study of unsupervised representation learning, or also under the context of few shot learning. In this sense, this project proposes the study of combinations of unsupervised and supervised architectures and their loss functions, investigating training strategies to allow finding general representations, not only for the training data domain, but also for alternative domains. (AU)