Pitt HomeFind PeopleContact Us

"Food for Thought" Graduate Colloquium Series

In conjunction with its regular Coffee Hour, the CS-GSO presents a periodic series of graduate colloquia, short 20- to 30-minute student-led presentations aimed at informing, engaging, and enriching the student research community in the Computer Science Department.

Fall 2014 (Term 2151)

3:00pm, SENSQ 6329
Missing Value Estimation for Hierarchical Time Series

Hierarchical time series (HTS) modeling is crucial and serves as the basis for business planning and management in many application areas such as manufacturing inventory, energy and traffic management. Time series used in these areas are always highly correlated and follow an organizational hierarchy. However, due to machine failures, network disturbances or simply human mistakes, HTS data frequently contain missing values across different hierarchical levels, which precludes a large class of techniques that require complete observational data input. In the literature, various techniques have been developed with great successes to deal with missing values in individual sequences or sparse low rank matrices. However, none of them satisfies the hierarchical consistency in HTS settings. In this paper, we estimate missing data in multivariate time series with a specific hierarchical structure. We develop a new iterative algorithm HTSImpute which (1) utilizes the temporal dependence information within each individual time series; (2) exploits the intra-relations between time series through hierarchy; (3) guarantees the satisfaction of hierarchical consistency constraints. The proposed method is evaluated on both synthetic and real-world traffic data sets, and the results demonstrate that our approach is better for HTS missing value imputation and outperforms alternative imputation methods in the HTS setting.

Terms

Top
Follow Pitt CS-GSO & SCI-GSO on Facebook

© 2009-2024 Pitt CS & SCI GSO