KarlRohe

@stat.wisc.edu

 

Broad Professional Interests

QUANTITATIVE INFERENCE

data collection, exploratory analysis, visualization, modeling, validation.

STATISTICAL MACHINE LEARNING
special attention to multivariate methods, clustering, and computationally tractable methods for model selection in
high-dimensions.    

APPLICATION TO COMPLEX SYSTEMS

e.g. social networks

 

V Vu, J Cho, J Lei, K Rohe. “Fantope Projection and Selection: A near-optimal convex relaxation of Sparse PCA”. NIPS 2013.  To appear.

T Qin and K Rohe.  “Regularized Spectral Clustering Under the Degree-Corrected Stochastic Blockmodel.”  NIPS 2013.  [pdf]

K Rohe and T Qin.  “The Blessing of Transitivity in Sparse and Stochastic Networks.”  [pdf]

J Jia and K Rohe.**  “Preconditioning to comply with the Irrepresentable Condition.”  [pdf]

K Rohe, T Qin, H Fan.  “The Highest Dimensional Stochastic Blockmodel with a Regularized Estimator.”  [pdf]

K Rohe and B Yu.  “Co-clustering for directed graphs; the stochastic co-blockmodel and a spectral algorithm.” Technical Report. [pdf]

K Rohe, S Chatterjee, and B Yu.  “Spectral clustering and the high-dimensional Stochastic Blockmodel.”  Annals of Statistics, 39(4):1878–1915, 2011. [pdf]

J Jia, K Rohe, and B Yu. “The Lasso under heteroskedasticity.” Statistica Sinica. [pdf]





** Authors contributed equally.

Publications

December.  I will speak at ERCIM in London.

December.  Both the paper on regularized spectral clustering and the Fantope paper will be at NIPS.

October 22nd.  I will speak about the Blessings paper at SAMSI’s workshop “Social Network Data: Collection and Analysis.

Spectral clustering applied to a sparse graph benfits from some regularization.  See my recent paper with Tai Qin here.

October 7th.  I’ll be visiting the University of Washington Department of Statistics to talk about the Blessings paper.

Transitivity, a common feature of empirical networks, can make clustering easy in sparse graphs.  In this new paper with Tai Qin we provide the first estimation results for local clustering.

August 4-8.  I will speak at JSM in Montreal.

July 23-25.  I will speak at the Duke Workshop on Sensing and Analysis of High-Dimensional Data.

April 5. I will speak at NYU Stern’s IOMS seminar series.

Feb 22.  I will speak at the Duke ECE department on preconditioning.

January 30 - Feb 1.  Great workshop at Eurandom.

November 14th and 15th 2012. I will be at Johns Hopkins University.  I will speak in the Applied Math and Stat Dept seminar on Thursday the 15th.

August 2012:  A simple preprocessing step can make the Lasso sign consistent under less restrictive assumptions.  See the details in this paper with J Jia.

Updates

(let me know if you will be nearby and would like to meet.)

photo courtesy of Frances Tong

This research is supported by DMS-1309998.