Minicurso: Anomaly Detection
Período de inscrição: até 27 de abril de 2016
Taxa de inscrição: R$ 5,00 - pago na Seção de Eventos do ICMC
Data de realização: 29 de Abril de 2016
Horário: das 10 às 12 horas
Local: Auditório Fernão Stella de Rodrigues Germano, ICMC
Professor: Dino Ienco
Short Class Description:
In this course I will introduce some basic concepts about the task of anomaly detection.
A quick overview about the families of methods will be presented and common techniques that belong to the state of the art of anomaly detection will be described.
At the same, I will depict the challenges and the current trend sin the field of outlier detection (subspace anomalies, graph anomalies, etc..).
In the second part of the talk I will discuss a recent approach we proposed [Ienco16], named SanDCat, that works in a semi-supervised way in order to detect anomalies given a training set composed by only "normal" examples. SanDCat is especially tailored to manage data represented by categorical and mixed (both numerical and categorical) attributes. In the talk I will also underline how the knowledge extracted by our anomaly detection method can be employed to explore and browse the dataset to characterize the highlighted anomalies.
Dino Ienco obtained his Ph.D. in Computer Science in 2010 at the University of Torino. Since September 2011 he obtained a permanent position as researcher at the Irstea Institute, Montpellier, France. His research interests are in the areas of data mining, data base and machine learning.
Considering Machine Learning approaches, he mainly focused on unsupervised (clustering and co-clustering) and semi-supervised (anomaly detection, active and positive unlabeled learning) analysis with application to spatio-temporal data with a major emphasis on remote sensing data. From a Data Base and Data Mining perspective, he is investigating graph-based data management approaches to model complex objects interactions ranging from indexing techniques to speed up data access to pattern mining extraction in order to mine useful knowledge from graphs.