Master Seminar:
HMMs: Universal tool?

The Hidden Markov Models (HMMs) are extremely powerful mathematical models which make it possible to model the characteristics of sequential signals (temporal or different). The use of the Hidden Markov Models spreads in very diverse fields such as:

• Automatic speech recognition

• Handwriting writing recognition

• Biometrics: verifying of speaker and writer

• Bio-informatic: research in sequences ADN, modeling of proteins

• Linguistics: modeling of text

• Image processing: recognition of forms and segmentations

• Climatology: forecast of precipitations

• Economy: modeling of "business cycles"

The objective of this seminar is to make a state of the art on the use of the Hidden Markov Models in some fields mentioned above and to show the forces and weaknesses of these models in the contexts of use.

Thus, it will be requested from each student to choose an applicability of the Hidden Markov Models among those listed above, of reading the corresponding articles, synthesizing them and finally to present them at the end of the seminar. Moreover, one ratio of 4-6 pages will have to synthesize the readings and to present the research tasks in the selected field.

The interested students must announce themselves by Jean Hennebert or Rolf Ingold since the number of students will be limited to 6-8 people (i.e. first registered voters).

When?

• Friday, 08 April, 14h:
  • Presentation of the seminar and overview [PDF]
  • Choice of the topics by the master-students

  
  

• Tuesday, 31 May, 14h:
  • Talks:
    • Andreas Schlapbach, Uni Bern: Using HMMs for Writer Identification
    • Tamas Varga, Uni Bern: Line Distortions for Handwriting Recognition
    • Roman Bertolami, Uni Bern: Multiple Handwritten Text Line Recognition System Derived from Specific Integration of a Language Model
    • Marcus Liwicki, Uni Bern: Processing Whiteboard Notes
    • Asmaa El Hannani, Uni Fribourg: Using Data-Driven Speech Segmentation for Speaker Verification

• Friday, 24 June, 14h:
  • Final student presentations.

References :

- Biosciences

• “Meta-MEME: Motif-based hidden Markov models of protein families”, W. N. Grundy, T. L. Bailey, M. E. Baker, Computer applications in the biosciences, 13(4), p. 397, aug 1997 [PDF]
• “Predicting Peptides That Bind to MHC Molecules Using Supervised Learning of Hidden Markov Models”, Hiroshi Mamitsuka, Proteins, 33(4), p. 460, 1998 [PDF]
• “Fold Recognition Using Predicted Secondary Structure Sequences and Hidden Markov Models of Protein Folds”, Di Francesco, Valentina , V. Geetha, Peter J. Munson, Proteins, , p. 123, 1997 supp1 [PDF]

- Climatology

• “A hidden Markov model for downscaling synoptic atmospheric patterns to precipitation amounts”, E. Bellone, J.P. Hughes, P. Guttorp, Climate research, 15(1), p. 1, may 2000 [PDF]
• “A non-homogeneous hidden Markov model for precipitation occurrence”, J.P. Hughes, P. Guttorp, S.P. Charles, Applied statistics, 48(1), p. 15, 1999 [PDF]

- Handwritting and text recognition and writer verification

• “Hidden Markov Models in Text Recognition”, J. Anigbogu, A. Belaid, International journal of pattern recognition and, 9(6), dec 1995 [PS]
• “Off-line cursive handwriting recognition using hidden Markov models”, H. Bunke, M. Roth, E. Schukat- Talamazzini, Pattern recognition, 28(9), 1995 [PS]

-  Writer verification

• “Using HMM-based recognizers for writer identification and verification”, A. Schlapbach and H Bunke, Proc. 9th Int. Workshop on Frontiers in Handwriting Recognition, 2004, pp. 167 - 172 [PDF]
• “Off-line handwriting identification using HMM-based recognizers”, A. Schlapbach and H. Bunke, Proc. 17th Int. Conference on Pattern Recognition, vol. II, 2004, pp. 654 - 658 [PDF]