| Grant number: | 13/20387-7 |
| Support Opportunities: | Scholarships abroad - Research |
| Start date: | March 01, 2014 |
| End date: | February 28, 2015 |
| Field of knowledge: | Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques |
| Principal Investigator: | João Paulo Papa |
| Grantee: | João Paulo Papa |
| Host Investigator: | David Cox |
| Host Institution: | Faculdade de Ciências (FC). Universidade Estadual Paulista (UNESP). Campus de Bauru. Bauru , SP, Brazil |
| Institution abroad: | Harvard University, Cambridge, United States |
| Associated research grant: | 09/16206-1 - New trends on optimum-path forest-based pattern recognition, AP.JP |
Abstract %Deep learning architectures have been extensively studied in the last years, since their philosophy consists into using a complex hierarchical structure for information learning and representation, being such representation analogous to the human neural processing. Such architectures are composed by several steps, which aim to use an image (in case of computer vision applications) in a filtering process using a filter bank for further information extraction and sampling. Then, this image is modified and forwarded to a new operation layer, in the same way as conducted in the previous one. At the final of the process, it is obtained a high dimensional description of this image, being such representation employed in a traditional classification process. The main problem concerns with the number of parameters of the hole process (filter number and sizes, sampling rate and classifiers' parameters, for instance), which are so called hyperparameters and they are fundamental for the success of the information extraction, representation and classification. Therefore, we can model the task of finding such parameters as an optimization problem. Since a few number of works have addressed evolutionary optimization techniques in this context, this post-doctorate research project aims to employ and evaluate such optimization techniques for hyperparameter optimization in deep learning architectures. Additionally, this work as the goal to evaluate the Optimum-Path Forest classifier, which was developed by the proposer of this project, in the context of data classification in deep learning, since OPF has never been applied for such purpose so far. (AU) | |
| News published in Agência FAPESP Newsletter about the scholarship: | |
| More itemsLess items | |
| TITULO | |
| Articles published in other media outlets ( ): | |
| More itemsLess items | |
| VEICULO: TITULO (DATA) | |
| VEICULO: TITULO (DATA) | |