Title: Dynamic tracking and screening in massive datastreams
Authors: Lilun Du - HKUST (Hong Kong) [presenting]
Abstract: In the modern era, technological advances have led to the emergence of an increasing number of applications requiring analysis of large-scale datasreams, that are consisted of multiple indefinitely long and time evolving sequences. Consequently, it is often necessary to develop statistical methodologies that perform inferential tasks in an online manner, and can continuously revise the model to reflect the current status of the underlying process. In particular, we are interested in constructing a large scale dynamic tracking and screening (DTS) procedure capable of rapidly identifying irregular individual streams whose behavioral patterns deviate from that of the majority. By fully exploiting the sequential feature of datastreams, we first develop a robust estimation approach under a framework of varying coefficient model. The procedure naturally accommodates unequally-spaced design points and updates estimates as new data arrive without the need to store an ever increasing data history. A data driven choice of an optimal smoothing parameter is accordingly proposed. Then, we suggest a new model-specification test tailored to the streaming environment. The resulting DTS scheme is able to adapt time varying structures appropriately, track changes in the underlying models, and hence maintain high identification accuracy in detecting irregular individuals.