Data depth, Multiterminal estimation, Applications
 
What's new?
Rebecka Jornsten, PhD.
Assistant Professor,
Department of Statistics
Rutgers University
110 Frelinghuysen Road,
Piscataway, NJ 08854, USA

Email: rebecka@stat.rutgers.edu
Phone: (732) 445-3145
FAX: (732) 445-3428
Office: 451 Hill Center

Introduction Papers and Presentations Other

[Overview | Extensions of the Band Depth | Clustering and Classification using Data Depth| Comprestimation]

  • The concept of Data Depth generalizes the median to a multivariate setting, and defines a center-outward ranking of observations. I am working on the development of clustering methodologies based on several notions of Data Depth, with application to the robust analysis of multivariate and functional data. Together with our recent post-doctoral visitor, Sara Lopez, I am investigating extensions of data depth to account for replicate measurements and measurement uncertainty.

  • Functional analysis via Extensions of the Band Depth Most depth measures that are applied to functional data are in fact invariant with respect to permutations of the data dimensions, and do not take the implicit ordering or structure of functional data into account. In a recent paper, we propose extensions of the Band Depth that better account for functional structure. In this work, we make a distinction between outliers that are "random" with respect to the data dimensions, and "shape outliers". We show that this approach provides better location estimates for functional data than previously proposed methods. We are currently working on the development of classification and clustering methods using these notions of depth.

  • Clustering and classification via the L1 Data Depth
    In this paper, I proposed a method to use the concept of data depth for clustering. The DDclust method can detect clusters of varying size and shape, and is fully non-parametric.
    I am currently working on extensions to include variable selection in the non-parametric clustering framework.

  • Comprestimation
    When data are compressed prior to statistical analysis information is lost. However, sometimes compression can work to our favor. Lossy compression tries to strike a balance between compressing "noise" and preserving "signal".
    Thus, compression can sometimes be viewed as a denoising procedure, and can in fact help with the identification of the relevant information hidden in the data. Of course, this assumes that the compression was "informed" in some sense, focused on preserving the important aspects of the data.
    This is the concept of Comprestimation. Under various compression schemes, informed and uninformed, can we determine what the estimation effiency gains/losses are?
    With Bin Yu, I wrote several papers on Comprestimation. These days, my student, Owen Martin, is working on Comprestimation in the context of regularized regression.
    		• Data Depth
  • Functional analysis via extensions of the Band Depth
    Sara Lopez-Pintado and Rebecka Jornsten
    To appear in IMS Lecture Notes -- Monograph Series, Volume 54 TITLE: Complex datasets and inverse problems: tomography, networks and beyond. Regina Liu, William Strawderman & Cun-Hui Zhang,, Editors

  • DDclust Clustering and Classification based on the L1 data depth,
    Rebecka Jornsten
    Journal of Multivariate Analysis Volume 90, Issue 1 , July 2004, Pages 67-89

  • Clustering based on Data Depth, Statistics seminar, BU
  • A Robust Clustering Method and Visualization Tool Based on Data Depth,
    Rebecka Jornsten, Yehuda Vardi and Cunhui Zhang
    Statistical data analysis based on the L1norm and related methods. Birkhauser 2002, Statistics for industry and technology. Y. Dodge editor.
  • Cluster validation via the Relative Data Depth

    Comprestimation

  • Multiterminal estimation - extensions and a geometric intepretation, Peer-reviewed extended abstract for ISIT 2002, Lausanne
  • Multiterminal estimation - extensions and a geometric intepretation slides from ISIT2002.
  • Insensitivity of Adaptive Quantization to Model Estimation in Wavelet Subband Coding
    Rebecka Jornsten and Bin Yu
    Technical report, UC Berkeley, Department of Statistics

    • Analysis of EEG data

  • A neurobiological theory of meaning in perception. Part 3. Multiple cortical areas synchronize without loss of local autonomy. Walter J Freeman, Gyongyi Gaal, Rebecka Jornsten International Journal of Bifurcation & Chaos (2003)
  • Visit The Hart Lab at Rutgers University.
  • Visit ebCTC, environmental bioinformatics and Computational Toxicology Center. This Consortium of UMDNJ, Rutgers University and Princeton University is funded by USEPA STAR Grant number GAD R 832721-010.