Novel approaches for fairness and transparency in machine learning problems

• Grant number: 20/10572-5
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: December 01, 2020
• End date: February 04, 2024
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: Leonardo Tomazeli Duarte
• Grantee: Guilherme Dean Pelegrina
• Host Institution: Faculdade de Ciências Aplicadas (FCA). Universidade Estadual de Campinas (UNICAMP). Limeira , SP, Brazil
• Associated research grant(s): 24/18794-8 - LVI Brazilian Symposium on Operational Research (SBPO 2024), AR.BR
• Associated scholarship(s): 21/11086-0 - Interpretability and fairness in machine learning: Capacity-based functions and interaction indices, BE.EP.PD

Abstract

Machine learning techniques have been used in the construction of automatic systems in order to deal with several practical problems. Examples include the applications in credit systems, which evaluate if an individual will lead to a possible default with respect to the received credit, or in judicial systems, which predict whether the defendant under trial may re-offend. In general, the purpose of such systems is to aid decision makers in their complex tasks, which can be difficult to be dealt with due to the large amount of available information or the users inherent biases. However, what is frequently observed in real situations is that some algorithms promotes discrimination against specific groups of individuals. Therefore, there is a need in the development of machine learning techniques that take into account characteristics such as fairness and transparency in the construction of the adopted system. The goals of this research project lies on this context. More precisely, the aforementioned concerns will be addressed both in the pre-processing step, by means of an approach based on principal component analysis, and in both training and classification steps, through the use of multi-objective optimization and Choquet integrals. It is worth mentioning that, by using the proposed methods, it will be possible to build automatic systems whose application does not promote ethical disparities with respect to the individuals under analysis. Moreover, our proposals are generalist, i.e., they can be used to deal with several problems in machine learning.

Scientific publications

(6)

(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)

PELEGRINA, GUILHERME DEAN; COUCEIRO, MIGUEL; DUARTE, LEONARDO TOMAZELI. A preprocessing Shapley value-based approach to detect relevant and disparity prone features in machine learning. PROCEEDINGS OF THE 2024 ACM CONFERENCE ON FAIRNESS, ACCOUNTABILITY, AND TRANSPARENCY, ACM FACCT 2024, v. N/A, p. 11-pg., 2024-01-01. (21/11086-0, 20/10572-5, 20/09838-0)

CAMPELLO, BETANIA SILVA CARNEIRO; PELEGRINA, GUILHERME DEAN; PELISSARI, RENATA; SUYAMA, RICARDO; DUARTE, LEONARDO TOMAZELI. Mitigating subjectivity and bias in AI development indices: A robust approach to redefining country rankings. EXPERT SYSTEMS WITH APPLICATIONS, v. 255, p. 13-pg., 2024-07-24. (23/04159-6, 20/10572-5, 20/09838-0)

PELEGRINA, GUILHERME DEAN; SIRAJ, SAJID; DUARTE, LEONARDO TOMAZELI; GRABISCH, MICHEL. Explaining contributions of features towards unfairness in classifiers: A novel threshold-dependent Shapley value-based approach. ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE, v. 138, p. 12-pg., 2024-10-05. (21/11086-0, 20/10572-5, 20/09838-0)

PELEGRINA, GUILHERME DEAN; DUARTE, LEONARDO TOMAZELI; GRABISCH, MICHEL. A k-additive Choquet integral-based approach to approximate the SHAP values for local interpretability in machine learning. ARTIFICIAL INTELLIGENCE, v. 325, p. 23-pg., 2023-09-28. (21/11086-0, 20/10572-5, 20/09838-0)

PELEGRINA, GUILHERME D.; DUARTE, LEONARDO T.; GRABISCH, MICHEL. Interpreting the Contribution of Sensors in Blind Source Extraction by Means of Shapley Values. IEEE SIGNAL PROCESSING LETTERS, v. 30, p. 5-pg., 2023-01-01. (20/09838-0, 20/10572-5, 21/11086-0)