Domain adaptation with minimal supervision in multimedia problems

• Grant number: 15/09169-3
• Support Opportunities: Scholarships in Brazil - Doctorate
• Effective date (Start): August 01, 2015
• Effective date (End): June 30, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Acordo de Cooperação: Coordination of Improvement of Higher Education Personnel (CAPES)
• Principal Investigator: Ricardo da Silva Torres
• Grantee: Luis Augusto Martins Pereira
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

A common assumption in the conventional machine learning setting is that the training set and the test set follow the same probability distribution. Unfortunately, in many real-world applications, data captured by different devices or under varied acquisition conditions may cause distribution mismatch between these sets. One way to address this problem is through domain adaptation - which can be seen as a transfer learning approach, since the goal is to adapt a model trained on a source domain to recognize new instances from a new target domain. A shortcoming of this methods, however, it is assumed (during the model building) a large amount of labeled data available in the source domain and, depending on the case, a small amount of labeled data available in the target domain. Nonetheless, annotating data is a costly process, mainly for the problems involving images or videos, whose data sets may contain millions of instances. In view of this context, our goal in this research project is to investigate and to propose domain adaptation methods that enable the use of minimal supervision - i.e., the use of restricted amount of labeled data in the source domain, which can reduce the need for large amount of labeled data. Among the contributions of this project is the application of the developed methods as solutions for domain adaptation problems in two Microsoft/FAPESP projects (#2013/50155-0 and #2013/50169-1), which seek, respectively, a better understanding of phenological and ecological processes in the face of climate changes. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

PEREIRA, LUIS A. M.; TORRES, RICARDO DA SILVA. Semi-supervised transfer subspace for domain adaptation. PATTERN RECOGNITION, v. 75, n. SI, p. 235-249, MAR 2018. (13/50155-0, 15/09169-3, 13/50169-1)

PEREIRA, LUIS A. M.; TORRES, RICARDO DA SILVA. Semi-supervised transfer subspace for domain adaptation. PATTERN RECOGNITION, v. 75, p. 15-pg., 2018-03-01. (13/50169-1, 15/09169-3, 13/50155-0)

PIRES, RAFAEL; LEVADA, ALEXANDRE L. M.; SOUZA, GUSTAVO B.; PEREIRA, LUIS A. M.; SANTOS, DANIEL F. S.; PAPA, JOAO P.; IEEE. A Robust Restricted Boltzmann Machine for Binary Image Denoising. 2017 30TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI), v. N/A, p. 7-pg., 2017-01-01. (14/12236-1, 16/19403-6, 15/09169-3)

Academic Publications

(References retrieved automatically from State of São Paulo Research Institutions)

PEREIRA, Luis Augusto Martins. Adaptação de domínio via aprendizado de subespaço e métodos de kernel. 2018. Doctoral Thesis - Universidade Estadual de Campinas (UNICAMP). Instituto de Computação Campinas, SP.