Universidade Federal do Ceará
Mestrado e Doutorado em Ciências da Computação

Título: Gaussian Process Latent Variable Models for Anomaly Detection: Implementation and Practical Considerations

Data: 27/08/2024
Horário: 16h00
Local:  Sala de Seminários – Bloco 952 e Google Meet (https://meet.google.com/gxk-xvfd-gzv)

Resumo:

Anomaly detection is a key task in diverse domains, including banking, healthcare, and cybersecurity, among others. Anomaly detection involves the identification of data points that exhibit behavior that is different from the typical behavior of the system, which may indicate the occurrence of an abnormal event. Precise and quick response and decision making rely heavily on the accurate and real-time detection of anomalies. Gaussian Processes (GPs) are a versatile and robust non-parametric machine learning method that can be employed for a range of tasks, such as regression, classification, and time-series modeling. One key benefit of GPs is their ability to offer a probabilistic framework for representing uncertainty. This makes them well-suited for situations when a comprehensive understanding of the underlying system is unattainable. Although GPs have significant promise for detecting anomalies, there is a scarcity of work that integrates these two topics. The absence of research in this area provides a chance to investigate the capabilities of GPs for anomaly detection and assess the benefits and constraints of this method. Utilizing a probabilistic framework in anomaly detection has numerous benefits. Firstly, it enables the measurement of uncertainty, which is essential for making decisions in practical situations. Furthermore, it offers a versatile structure for representing intricate connections between inputs and outputs, rendering it well-suited for managing data with multiple dimensions and non-linear characteristics. This master dissertation seeks to investigate the application of Gaussian Processes for anomaly detection and to conduct a thorough evaluation of this methodology. The objective is to enhance the existing body of knowledge by addressing the deficiency in the application of Gaussian Processes (GPs) for anomaly detection and to showcase the benefits and constraints of this method. The findings of this study will enhance comprehension of the capacities of general practitioners (GPs) in detecting anomalies, and will facilitate the identification of future research avenues in this domain.

Banca examinadora:

• Prof. Dr. César Lincoln Cavalcante Mattos (MDCC/UFC – Orientador)
• Prof. Dr. João Paulo do Vale Madeiro (MDCC/UFC)
• Prof. Dr. João Paulo Pordeus Gomes (UFC)