OUTLIER AND ATTACKER RESILIENT METHODS BASED ON RATING AND REPUTATION SYSTEM

• Autor(a) principal: Matias, Pedro Miguel Pereira
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/160632
• topic: Outlier Detection; Isolation Forest; Local Factor Outlier; One Class Support Vector Machines; Minimum Covariance Determinant; Rating and Reputation; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

A key component in an automatic surveillance system that can receive crowd-sourced data, such as an early forest fire detection system, must consider the possibility of corrupted data and also attacks on the processors in the network running the estimation task. In both cases, there is the need to introduce some process to decide when to remove a specific value from the computations. In this thesis, we study using reputation and rating metrics to construct an algorithm that is resilient to erroneous data and attacks in linear dynamical systems and compare it against traditional methods to remove outliers. It is shown in simulation that the presented methods have performance comparing or surpassing the traditional methods, which is an interesting outcome that reinforces the importance of the literature on rating and reputation use for resilient consensus and distributed optimization.