Pattern Recognition in Complex Networks using Distance Transform

• Grant number: 19/03277-0
• Support type: Scholarships abroad - Research Internship - Doctorate
• Effective date (Start): June 01, 2019
• Effective date (End): November 30, 2019
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal researcher: Odemir Martinez Bruno
• Grantee: Lucas Correia Ribas
• Supervisor abroad: Antoine Manzanera
• Home Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Research place: Ecole Nationale Supérieure des Techniques Avancées (ENSTA), France
• Associated to the scholarship: 16/23763-8 - Modeling and analysis of complex networks for computer vision, BP.DR

Abstract

Complex network is being increasingly used due to its flexibility and ability to represent and analyze any discrete system and many types of data. In this sense, pattern recognition and complex networks emerge as an important alternative in data science.Pattern recognition in networks aims the feature extraction to classify large-scale networks into several classes, instead of focusing on the topology properties of an isolated network. There is a huge literature on feature extraction and data classification based on regular networks (e.g. convolutional). However, in spite of significant progress achieved, there is an increasing demand for more general and sophisticated methods, in particular in the case of highly irregular or sparse data. The aim of this proposal is the development of methods for the characterization of network topology, thereby giving particular attention to networks that model image and video. To achieve this, in this proposal we intend to investigate different ways to obtain features from networks using the distance transform. More precisely, the distance transform will be applied in the network and information about the obtained distances (or labels) will be used to characterize different network topologies in order to perform classification tasks in the context of pattern recognition.In this research project, distance transforms are then considered as the fundamental tool to: (1)~extract significant features from data distributed over any kind of network, and (2)~allow a complex (irregular, sparse, multi-layered, etc) network to be processed like a regular network.Development of new methods for pattern recognition and analysis of complex networks is an important stage of the PhD project of the student.Therefore, we aim to contribute with methods for network characterization using statistical metrics and learned features (neural network) from the information of the distance transform on the network. The developed methods will also contribute to computer vision field and data science with new methods for feature extraction and pattern recognition.

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)

RIBAS, LUCAS C.; DE MESQUITA SA JUNIOR, JARBAS JOACI; MANZANERA, ANTOINE; BRUNO, ODEMIR M.. Learning graph representation with Randomized Neural Network for dynamic texture classification. APPLIED SOFT COMPUTING, v. 114, JAN 2022. (16/23763-8, 18/22214-6, 16/18809-9, 14/08026-1, 19/03277-0)