Advanced search
Start date
Betweenand

Applications of multidimensional data processing using tensor methods

Grant number: 14/23936-4
Support type:Scholarships in Brazil - Post-Doctorate
Effective date (Start): April 01, 2015
Effective date (End): May 31, 2019
Field of knowledge:Engineering - Electrical Engineering
Cooperation agreement: Coordination of Improvement of Higher Education Personnel (CAPES)
Principal Investigator:João Marcos Travassos Romano
Grantee:Michele Nazareth da Costa
Home Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
Associated scholarship(s):16/14142-0 - Tensor networks and deep learning for large scale machine learning and signal processing problems, BE.EP.PD

Abstract

Several recent research topics related to the information processing are characterized by an increasing amount of data. As a consequence of this increase, the processing of these data tends to require more sophisticated methods than those classically employed in matrix approaches. Thus, the tensor approach is becoming a natural and powerful tool in this context, as it ensures that all data and intrinsic interactions are taking into account in the processing. Tensor methods can be developed and proposed from extensions of classical methods of matrix factorization and also from well-known tensor models, such as PARAFAC/CANDECOMP and Tucker.In her thesis, the candidate proposed a transmission system based on a Tensor Space-Time (TST) coding, with resources of antenna allocation, for MIMO wireless communication systems. The strategy adopted to improve the performance consisted of the joint use of different diversities inherent to the received signals. A new tensor decomposition was proposed by the candidate from which the signals received by multiple antennas were represented. Despite the promising results presented in her thesis, there are some aspects to be further investigated regarding the study of optimal allocation structure, optimization of the receivers, and analysis of the multiuser case. These issues compose the first part of this project, aimed at the application in Telecommunications. In contrast, it is also noticed that several digital signal processing techniques which are typically applied in Telecommunications context have been increasingly employed in the processing of seismic signals. Thus, the second part of this project is related to this application, in which the idea is to explore the tensorial formalism. Therefore, different topics of research will be investigated based on Principal Component Analysis (PCA) for separation of seismic waves, noise attenuation, and reconstruction of corrupted or incomplete data. The tensor tools will be employed for modeling and processing of seismic multidimensional data, in order to provide a better use of the information underlying seismic acquisitions, and also to aid the image generation of the subsurface volume of the region to be explored. (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)
DA COSTA, MICHELE NAZARETH; FAVIER, GERARD; ROMANO, JOAO MARCOS T. Tensor modelling of MIMO communication systems with performance analysis and Kronecker receivers. Signal Processing, v. 145, p. 304-316, APR 2018. Web of Science Citations: 8.
VIGNERON, VINCENT; KODEWITZ, ANDREAS; DA COSTA, MICHELE NAZARETH; TOME, ANA MARIA; LANGLANG, ELMAR. Non-negative sub-tensor ensemble factorization (NsTEF) algorithm. A new incremental tensor factorization for large data sets. Signal Processing, v. 144, p. 77-86, MAR 2018. Web of Science Citations: 1.
DANTAS, CASSIO FRAGA; DA COSTA, MICHELE N.; LOPES, RENATO DA ROCHA. Learning Dictionaries as a Sum of Kronecker Products. IEEE SIGNAL PROCESSING LETTERS, v. 24, n. 5, p. 559-563, MAY 2017. Web of Science Citations: 9.

Please report errors in scientific publications list by writing to: cdi@fapesp.br.