Antisparsity and Equidity in signal processing: from blind source separation to fairness machine learning

• Grant number: 19/20899-4
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: July 01, 2020
• End date: March 10, 2024
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: João Marcos Travassos Romano
• Grantee: Renan Del Buono Brotto
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated scholarship(s): 22/04237-4 - Antisparsity and equity: from blind source separation to fair machine learning, BE.EP.DR

Abstract

We define a signal, in a broad sense, as an entity that carries information. To better analyze the information of interest, a range of techniques have been developed, especially the ones based on signal decomposition. These techniques are used according to the features we want to extract. Formally, they are implemented by means of matricial or tensorial decomposition. In our MSc research (process number 2017/13025-2), we treated the Blind Source Separation problem, which is based on signal decomposition methods. Particularly, we have studied the techniques based on Independent Component Analysis (ICA), which supposes a statistical independence between the sources, and Sparse Component Analysis (SCA), that explores the concentration of information in a few samples. We have also introduced the notion of Antisparse Blind Source Separation, a new concept for the field, up to our best knowledge. In a dual matter with the sparse case, an antisparse signal tends to distribute the information equally between its samples. In our first studies, we have demonstrated that the antisparsity can be used as a prior information to the BSS problem, considering instant, linear, determined mixtures generated from independent sources. As a natural consequence of these first results, it's of great interest to investigate how the antisparsity can be employed in other scenarios of BSS, as with statistical dependent sources, the subdetermined case and the case of convolutive mixtures. Another very interesting property of the antisparsity is its relation with the fairness concept, which has gained much attention recently, mainly in Machine Learning. The Machine Learning techniques have been applied in many fields, including problems with socio-economic impact. In these problems, the statistical model can incorporate biased features of the dataset, leading, possibly, to unfair decisions. Therefore, is of great relevance to incorporate into the data based learning, mechanisms that promote fairness. Our work proposal is to achieve fairness in two steps: in the first one, we preprocess the dataset, trying to reduce the influence of the discriminative attributes over the others. In the second step, we search to reduce the influence of the discriminatory attributes over the problem classes, investigating mechanisms for fairness model adjustment. As is known in the literature, there is a loss of accuracy as we improve fairness in classification. So it is of great importance to investigate theoretical limits for the compromise between accuracy and fairness. For this, the techniques of statistical signal processing and of the Information Theory have a great solution potential. (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)

PELEGRINA, GUILHERME D.; BROTTO, RENAN D. B.; DUARTE, LEONARDO T.; ATTUX, ROMIS; ROMANO, JOAO M. T.; IEEE. Analysis of Trade-offs in Fair Principal Component Analysis Based on Multi-objective Optimization. 2022 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), v. N/A, p. 8-pg., 2022-01-01. (20/09838-0, 19/20899-4, 20/01089-9, 20/10572-5, 21/11086-0)

JUNIOR, MAURO LUIZ BRANDAO; LIMA, VICTOR CARNEIRO; BROTTO, RENAN DEL BUONO; ALVIM, JOAO RABELLO; PEREIRA, THOMAS ANTONIO PORTUGAL; LOPES, RENATO DA ROCHA; ROMANO, JOAO MARCOS TRAVASSOS; NOSE-FILHO, KENJI. IMPROVING IMAGE DEBLURRING. INVERSE PROBLEMS AND IMAGING, v. N/A, p. 18-pg., 2022-11-22. (19/20899-4, 20/09838-0)

NOSE-FILHO, KENJI; LOPES, RENATO; BROTTO, RENAN D. B.; SENNA, THONIA C.; ROMANO, JOAO M. T.. Algorithms for Sparse Multichannel Blind Deconvolution. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, v. 61, p. 7-pg., 2023-01-01. (19/20899-4, 20/09838-0)

Academic Publications

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

BROTTO, Renan Del Buono. Inverse problems applied to blind source separation and fair machine learning. 2024. Doctoral Thesis - Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação Campinas, SP.