Abstract
This paper describes an efficient and effective design of Robust Spatio-Temporal Prediction based on Student's t distribution, namely, St-RSTP, to provide estimations based on observations over spatio-temporal neighbors. The proposed St-RSTP is more resilient to outliers or other small departures from model assumptions than its ancestor, the Spatio-Temporal Random Effects (STRE) model. STRE is a state-of-the-art statistical model with linear order complexity for large scale processing. However, it assumes Gaussian observations, which has the well-known limitation of non-robustness. In our St-RSTP design, the measurement error follows Student's t distribution, instead of a traditional Gaussian distribution. This design reduces the influence of outliers, improves prediction quality, and keeps the problem analytically intractable. We propose a novel approximate inference approach, which approximates the model into the form that separates the high dimensional latent variables into groups, and then estimates the posterior distributions of different groups of variables separately in the framework of Expectation Propagation. As a good property, our approximate approach degeneralizes to the standard STRE based prediction, when the degree of freedom of the Student's t distribution is set to infinite. Extensive experimental evaluations based on both simulation and real-life data sets demonstrated the robustness and the efficiency of our Student-t prediction model. The proposed approach provides critical functionality for stochastic processes on spatio-temporal data.
| Original language | English (US) |
|---|---|
| Title of host publication | Proceedings - IEEE International Conference on Data Mining, ICDM |
| Pages | 151-160 |
| Number of pages | 10 |
| DOIs | |
| State | Published - 2012 |
| Externally published | Yes |
| Event | 12th IEEE International Conference on Data Mining, ICDM 2012 - Brussels, Belgium Duration: Dec 10 2012 → Dec 13 2012 |
Other
| Other | 12th IEEE International Conference on Data Mining, ICDM 2012 |
|---|---|
| Country | Belgium |
| City | Brussels |
| Period | 12/10/12 → 12/13/12 |
Fingerprint
Keywords
- Expectation propagation
- Spatio-temporal process
- Student's t distribution
ASJC Scopus subject areas
- Engineering(all)
Cite this
Student-t based Robust Spatio-Temporal Prediction. / Chen, Yang; Chen, Feng; Dai, Jing; Clancy, T. Charles; Wu, Yao-jan.
Proceedings - IEEE International Conference on Data Mining, ICDM. 2012. p. 151-160 6413907.Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
}
TY - GEN
T1 - Student-t based Robust Spatio-Temporal Prediction
AU - Chen, Yang
AU - Chen, Feng
AU - Dai, Jing
AU - Clancy, T. Charles
AU - Wu, Yao-jan
PY - 2012
Y1 - 2012
N2 - This paper describes an efficient and effective design of Robust Spatio-Temporal Prediction based on Student's t distribution, namely, St-RSTP, to provide estimations based on observations over spatio-temporal neighbors. The proposed St-RSTP is more resilient to outliers or other small departures from model assumptions than its ancestor, the Spatio-Temporal Random Effects (STRE) model. STRE is a state-of-the-art statistical model with linear order complexity for large scale processing. However, it assumes Gaussian observations, which has the well-known limitation of non-robustness. In our St-RSTP design, the measurement error follows Student's t distribution, instead of a traditional Gaussian distribution. This design reduces the influence of outliers, improves prediction quality, and keeps the problem analytically intractable. We propose a novel approximate inference approach, which approximates the model into the form that separates the high dimensional latent variables into groups, and then estimates the posterior distributions of different groups of variables separately in the framework of Expectation Propagation. As a good property, our approximate approach degeneralizes to the standard STRE based prediction, when the degree of freedom of the Student's t distribution is set to infinite. Extensive experimental evaluations based on both simulation and real-life data sets demonstrated the robustness and the efficiency of our Student-t prediction model. The proposed approach provides critical functionality for stochastic processes on spatio-temporal data.
AB - This paper describes an efficient and effective design of Robust Spatio-Temporal Prediction based on Student's t distribution, namely, St-RSTP, to provide estimations based on observations over spatio-temporal neighbors. The proposed St-RSTP is more resilient to outliers or other small departures from model assumptions than its ancestor, the Spatio-Temporal Random Effects (STRE) model. STRE is a state-of-the-art statistical model with linear order complexity for large scale processing. However, it assumes Gaussian observations, which has the well-known limitation of non-robustness. In our St-RSTP design, the measurement error follows Student's t distribution, instead of a traditional Gaussian distribution. This design reduces the influence of outliers, improves prediction quality, and keeps the problem analytically intractable. We propose a novel approximate inference approach, which approximates the model into the form that separates the high dimensional latent variables into groups, and then estimates the posterior distributions of different groups of variables separately in the framework of Expectation Propagation. As a good property, our approximate approach degeneralizes to the standard STRE based prediction, when the degree of freedom of the Student's t distribution is set to infinite. Extensive experimental evaluations based on both simulation and real-life data sets demonstrated the robustness and the efficiency of our Student-t prediction model. The proposed approach provides critical functionality for stochastic processes on spatio-temporal data.
KW - Expectation propagation
KW - Spatio-temporal process
KW - Student's t distribution
UR - http://www.scopus.com/inward/record.url?scp=84874025240&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84874025240&partnerID=8YFLogxK
U2 - 10.1109/ICDM.2012.135
DO - 10.1109/ICDM.2012.135
M3 - Conference contribution
AN - SCOPUS:84874025240
SN - 9780769549057
SP - 151
EP - 160
BT - Proceedings - IEEE International Conference on Data Mining, ICDM
ER -