A0560
Title: Learning under commission and omission event outliers
Authors: Guanhua Fang - Fudan University (China) [presenting]
Yuecheng Zhang - Fudan University (China)
Wen Yu - Fudan University (China)
Abstract: Event stream is an important data format in real life. The events are usually expected to follow some regular patterns over time. However, the patterns could be contaminated by unexpected absences or occurrences of events. The temporal point process framework is adopted for learning event streams, and a simple-but-effective method is provided to deal with both commission and omission event outliers. In particular, a novel weight function is introduced to dynamically adjust the importance of each observed event so that the final estimator could offer multiple statistical merits. The proposed method is compared with the vanilla one in two downstream tasks: Classification problems and change point detection problems. Both theoretical and numerical results confirm the effectiveness of our new approach. To best of knowledge, the method is the first one to provably handle both commission and omission outliers simultaneously.
