Fei Liu's Homepage

Ph.D.  in statistics
Advisor: James Berger
Insitute of Statistics and Decision Sciences



Research Interests:

Functional data analysis and Bayesian nonparametric statistics; computer model validation; spatial statistics; spatio-temporal models; Bayesian Hierarchical modeling, Multiple testing; data mining.

Publications / work in progress:
  • Bayarri, M. J., Berger, J. O., Cafeo, J., Garcia-Donato, G., Liu, F., Palomo, Parthasarathy, R.J., Paulo, R., Sacks, J., Walsh, D. (2006) Computer Model Validation with Functional Output. Annals of Statistics. To appear.
  • G. Garcia-Donato, F. Liu, and J. Palomo. (2004). Documentation for SAVE-2 methodology and software. Niss Tech. Rep.
  • F. Liu, M. J. Bayarri, J. Berger, R. Paulo, J. Sacks. (2006). A Bayesian Analysis of the Thermal Challenge Problem. Manuscript.
  • F. Liu, L. Zhang, M. West (2006). Bayesian Dynamic Model for Complex Computer Systems. Manuscript.
  • F. Liu, M. J. Bayarri, J. Berger. (2006). Modularization of the Bayesian Analysis for Random Effect Models. Work in progress.
  • F. Liu, M. J. Bayarri, J. Berger.(2006). A Multi-scale Bayesian Functional Data Analysis. Work in progress.
  • F. Liu, J. Berger. (2006). Bayesian Reconstruction of nonlinear dynamic systems. Work in progress.
  • F. Liu, F. Liang, W. Jang. (2006). A Bayesian method for partially-paired high dimensional data, with application to unpaired micro-array data. Work in progress.
Current Research:

My thesis, Bayesian Functional Data Analysis for Computer Model Validation, has developed the Bayesian space-time models for four types of functional data, arsing in the context of computer model validation. These methods utilize approaches such as Bayesian nonparametric statistics, Bayesian dynamic linear modeling, and Wavelet analysis. Besides the issues of functional data, these methods take into account other issues such as building emulators for expensive computer models, measuring discrepancy between the computer model output and the reality, and solving the statistical inverse problem for the unknown inputs associated with the field experiments.  The methods are,

  • Gaussian Stochastic Process -- The test bed case study for this model is: the Thermal Computer Model Challenge Problem. Talk slide on this methodology is here.
  • Wavelet analysis with Gaussian Stochastic Process method -- The test bed case study is a real commercial simulator (confidential). Talk slide on this methodology is  here  and  a  concise  version  here.
  • Wavelet PCA with Gaussian Stochastic Process method -- This is an on-going research, extending the Wavelet method to eigen-basis representation.
  •  multivariate Dynamic Linear Model (DLM) -- This method utilize the Bayesian dynamic linear model techniques to build emulators for the  simulators. I am trying to apply this method to CMAQ -- an air quality model used by EPA. Slide on this methodology is here.

I also collaborated with Professor Liang and Professor Jang on Bayesian analysis of partially paired data in high dimensional space. Our method extends Professor Berger's work on Bayesian Multiple Testing to cases with partially paired data. We compare this method with the FDR
adjustment approach by Professor Benjamini and Professor Hochberg. This model has been applied to a cDNA micro-array data set. The work was presented at the Case Studies in Bayesian Statistics, 2005 (slide). Here is a later version presented at SAMSI, 2006.

I am currently participating the SAMSI program on  Development, Assessment and Utilization of Complex Computer Models.
There are many interesting research topics and collabration opportunities. Of my particular interest are the methodology working group (I am in charge of this webpage :-) ), the Air Quanlity working group, and the Terrestrial working group