Selected projects

Hierarchical Dirichlet Process Hidden Markov Trees

Joint work with Erik B. Sudderth and Michael I. Jordan.

We developed a nonparametric Bayesian modeling framework in which data is generated via tree-structured probabilistic graphical models with latent variables, whose complexity is determined in a data-driven and -adaptive way. Tree-structured graphical models have natural ties, for example to multiscale data. The hierarchical Dirichlet process hidden Markov tree (HDP-HMT) framework extended prior work on, e.g., hidden Markov trees [CNB98], providing a principled and practical approach to determine the amounts of latent states, automatically via the use of hierarchical Dirichlet process [TJBB06]. We developed methodology for effective learning and inference with the framework, and explored applications to computational vision tasks, producing state-of-the-art results (at the time).

Conference and workshop publications and their related presentation materials

• Transfer Denoising with Hierarchical Dirichlet Process Hidden Markov Trees
  Jyri J. Kivinen, Erik B. Sudderth, and Michael I. Jordan
  In Nonparametric Bayes Workshop (NPBayes 2008) at ICML/UAI/COLT 2008, July 2008, Helsinki, Finland
  Poster, Spotlight

• Learning Multiscale Representations of Natural Scenes Using Dirichlet Processes
  Jyri J. Kivinen, Erik B. Sudderth, and Michael I. Jordan
  In Proc., IEEE International Conference on Computer Vision (ICCV), Oct. 2007, Rio de Janeiro, Brazil
  Poster

• Image Denoising with Nonparametric Hidden Markov Trees
  Jyri J. Kivinen, Erik B. Sudderth, and Michael I. Jordan
  In Proc., IEEE International Conference on Image Processing (ICIP), Sep. 2007, San Antonio, Texas, USA
  Slides

Drafts

• Hierarhical Dirichlet Process Hidden Markov Trees for Multiscale Image Analysis
  Jyri J. Kivinen, Erik B. Sudderth, and Michael I. Jordan

Selected other resources

• Multiscale Image Modeling and Analysis with Hierarchical Dirichlet Process Hidden Markov Trees
  Jyri J. Kivinen
  Invited talk at Information and Computer Science Forum, Helsinki University of Technology, 21 Aug. 2009

• Infinite Hidden Markov Trees
  Erik B. Sudderth
  Applied Bayesian Nonparametrics, Tutorial Course at CVPR 2012, 17 June 2012

References

CNB98

M. S. Crouse, R. D. Nowak, and R. G. Baraniuk. Wavelet–based statistical signal processing using hidden Markov models. IEEE Trans. Sig. Proc., 46(4):886–902, 1998.

TJBB06

Y. W. Teh, M. I. Jordan, M. J. Beal, and D. M. Blei. Hierarchical Dirichlet processes. J. Amer. Stat. Assoc., 101(476):1566–1581, 2006.