31 dez 2021
14:00 Master's Defense Fully distance
Theme
Neural Architecture Search Analysis for Object Detection
Student
Cicera Vanessa Marques Sampaio Sidrim
Advisor / Teacher
Sandra Eliza Fontes from Avila
Brief summary
Neural Architecture Search (NAS) is the automation of the process of building deep neural network architectures. The main objective is to reduce manual efforts and the time spent on this step by automatically generating an appropriate architecture for a given dataset. NAS was proposed in 2017 and, since then, several solutions have been introduced seeking strategies that require less computational power and expanding its performance to other tasks in the area of Computer Vision. However, we are still far from this reality where efforts and costs are reduced and the results are surprising. In this dissertation, we investigate the literature in order to identify NAS-based solutions for object detection task, better understanding how advanced and promising is the use of these networks. For this, we took into account a less robust infrastructure and a dataset with a lower amount of images compared to the use reported in the literature, bringing a more realistic view of the use of NAS, presenting the difficulties faced and justifying why we are still far away. of this reality.
Examination Board
Headlines:
| Sandra Eliza Fontes de Avila | IC / UNICAMP |
| Aurea Rossy Soriano Vargas | IC / UNICAMP |
| Levy Boccato | FEEC / UNICAMP |
Substitutes:
| Paula Dornhofer Paro Costa | FEEC / UNICAMP |
| Hélio Pedrini | IC / UNICAMP |