Introduction - On Statistical Pattern Recognition in Independent Component Analysis Mixture Modelling - page 24

relationship between ceramic physical properties, ultrasound propagation, and
method employed to manufacture the pieces is included.
In the second application of Chap. 6 , the ICA mixture algorithm is applied as a
non-parametric one-ICA algorithm for BSS. The goals are to evaluate the resto-
ration consolidation and to detect interfaces in a wall of a historical dome of a
Basilica using ultrasounds. The measured signals contain the contribution of the
injected ultrasonic pulse buried in backscattering grain noise plus the contribution
of sinusoidal phenomena. The sources and mixture matrix extracted by ICA allow
these contributions to be separated. The recovered sinusoidal sources characterize
the resonance phenomenon of multiple reflections of the ultrasonic pulse at non-
consolidated zones and instrument interferences.
Chapter 7 presents two different applications. First, a new model for sequential
pattern recognition based on HMM and ICAMM is proposed. The model is called
SICAMM. It is applied to the first application of Chap. 7 , which is the detection of
micro arousals caused by apnea during the night. These abrupt changes are reg-
istered in a diagram called hypnogram, which shows the transitions between the
sleep stages. Long EEG records from apnea patients measured during sleep are
analyzed. Two sleep stages are classified: wake and sleep. The classification
obtained by SICAMM outperforms the ICAMM classification showing accuracy
for detection of episodes of wakefulness during sleep. The second application of
Chap. 7 is webmining from a huge amount of historical web log data from
e-learning activities. The ICA mixture algorithm is configured to estimate the
parameters for only one ICA. The application consists of the detection of student
learning styles based on a known educational framework. The mixture matrix
obtained by ICA demonstrates the relation between e-learning style dimensions
and e-learning web activities leading to the detection of student learning styles.
Finally, Chap. 8 ends the thesis with the conclusions and findings, discussion of
the open relevant subjects, and discussion about the future directions of research.
References
1. C.J.D.M. Verhagen, Some general remarks about pattern recognition: its definition; its
relation with other disciplines; a literature survey. Pattern Recognit. 7(3), 109-116 (1975)
2. J.C. Bezdek, S.K. Pal, Fuzzy Models for Pattern Recognition: Methods That Search for
Structures in Data (IEEE press, New York, 1992)
3. H.A. Simon, Science Seeks Parsimony, Not Simplicity: Searching for Pattern In Phenomena.
in Simplicity, Inference and Modelling (Keeping it Sophisticatedly Simple), ed. by A. Zellner,
H.A. Keuzenkamp, M. McAleer (Cambridge University Press, Cambridge, 2004)
4. V. Castelli, T.M. Cover, The relative value of labelled and unlabelled samples in pattern
recognition with an unknown mixing parameter. IEEE Trans. Inf. Theory 42(6), 2102-2117
(1996)
5. I. Cohen, F.G. Cozman, N. Sebe, M.C. Cirelo, T.S. Huang, Semisupervised learning of
classifiers: theory, algorithms, and their application to human-computer interaction. IEEE
Trans. Pattern Anal. Mach. Learn. 26(12), 1553-1567 (2004)
6. O. Chapelle, B. Schölkopf, A Semi-supervised Learning (MIT Press, Zien, 2006)