Berkant Savas

Postdoctoral scholar at UT Austin

Research interests

My research background is in numerical linear and multilinear algebra, and in particular related to problems from data mining and pattern recognition applications. I earned my PhD in Scientific Computing from Linkoping University (Sweden) under the supervision of professor Lars Eldén. PhD thesis title: Algorithms in Data Mining using Matrix and Tensor Methods. Currently I am postdoc at the Data Mining Lab led by professor Inderjit Dhillon and pursuing research in a few different areas:
(1) Large scale computations for graphs and network problems, e.g. link prediction in dynamic networks, incorporation of multiple sources of information for link prediction and group recommendation.
(2) Stochastic methods for large scale low rank matrix approximation.
(3) Fast large scale eigenvector and singular vector computations.
(4) Computational methods and algorithms for tensors and tensor decompositions. In particular methods for low multilinear rank approximation of tensors. Most resent approach involves tensor Krylov methods.
(5) Optimization methods for problems defined on Grassmann manifolds.

Published articles, drafts and reports

Perturbation theory and optimality conditions for the best multilinear rank approximation of a tensor. Submitted to SIAM Journal on Matrix Analysis and Applications (SIMAX), 2011.

Clustered low rank approximation of graphs in information science applications. To appear in SIAM Data Mining Conference, 2011.

Scalable Affiliation Recommendation using Auxiliary Networks. To appear in ACM Transactions on Intelligent Systems and Technology.

Fast and accurate low rank approximation of massive graphs. Technical report TR-10-18, Department of Computer, University of Texas at Austin.

Clustered embedding of massive online social networks. In preparation.

Supervised link prediction using multiple sources. Proceedings of the IEEE International Conference on Data Mining (ICDM), 2010, pp. 923-928. BibTeX (email me for a copy)

Supervised link prediction using multiple sources (long version). Technical report TR-10-35, Department of Computer Science, The University of Texas at Austin, 2010. BibTeX

Krylov-type methods for tensor computations, Submitted for publication to Linear Algebra and its Applications. BibTeX

Krylov subspace methods for tensor computations, Technical report LiTH-MAT-R-2009-02-SE. BibTeX

Quasi-Newton methods on Grassmannians and multilinear approximations of tensors,
SIAM Journal on Scientific Computing, Volume 32, Number 6, pp. 3352-3393 (2010). BibTeX

A Newton-Grassmann method for computing the best multilinear rank-(r_1, r_2, r_3) approximation of a tensor,
SIAM Journal on Matrix Analysis and Applications, Volume 31, Issue 2, pp. 248-271 (2009). BibTeX

Handwritten digit classification using higher order singular value decomposition,
Pattern Recognition, Volume 40, Issue 3 , March 2007, Pages 993-1003. BibTeX

Dimensionality reduction and volume minimization - generalization of the determinant minimization criterion for reduced rank regression problems,
Linear Algebra and its Applications, Volume 418, Issue 1 , 1 October 2006, Pages 201-214. BibTeX

Rank reduction and volume minimization approach to state-space subspace system identification,
Signal Processing, Volume 86, Issue 11, November 2006, Pages 3275-3285. BibTeX

The maximum likelihood estimate in reduced rank regression,
Numerical Linear Algebra and Applications, Volume 12, Issue 8, Pages 731 - 741. BibTeX

Upcoming conferences and workshops

Attended conferences and workshops

My address at UT Austin:
Institute for Computational Engineering and Sciences
The University of Texas at Austin
1 University Station C0200, ACES 4.102
Austin, TX 78712
Office phone: 1-512-471-0026

Additional info can be found here .