![](bg6.png)
ent is chosen due to its advantages over first degree
optimization scheme like steepest descent and easy
implementation. We test the OCG-LDA algorithm
for various UCI datasets to demonstrate its classifi-
cation performance, average time and average itera-
tions. OCG-LDA clearly outperforms the L-2 LDA
and SD-LDA but has comparable performance with
the L-1 version of least square algorithms. However
the proposed methodology is simple and easy to im-
plement and is a good alternative to other algorithms
in building a robust model for classification. As from
No Free Lunch theorem, no single classification algo-
rithm can outperform any other algorithm when per-
formance is analyzed over many classification dataset.
In conclusion, OCG-LDA can be used as a basic clas-
sifier unit in a multi stage classification scheme.
7 FUTURE WORK
The OCG-LDA methodology is an evident advance-
ment in the L
1
family of LDA subspace algorithms.
As a part of future direction, a multiple optimal learn-
ing factor scheme based on the Gaussian Newton ap-
proximation (Malalur and Manry, 2010) can be in-
vestigated. Recently, the author (Cai et al., 2011)
have proposed an efficient partial Hessian calculation
that does not involves inversion and is successfully
applied on Radial basis function neural networks.
Therefore a study can be conducted to foray into the
second order algorithms using regularization parame-
ter.
REFERENCES
Bell, A. and sejnowski, T. (1995). An information-
maximization approach to blind separation and blind
deconvolution. Neural Computation, 7.
Cai, X., K.Tyagi, and Manry, M. (2011). An optimal con-
struction and training of second order rbf network for
approximation and illumination invariant image seg-
mentation.
Cao, L., Chua, K., Chong, W., Lee, H., and Gu, Q. (2003).
A comparison of pca, kpca and ica for dimensionality
reduction in support vector machine. Neurocomput-
ing, 55.
Chong, E. and Stanislaw, Z. (2013). An introduction to op-
timization. Wiley, USA, 3rd edition.
Claerbout, J. F. and Muir, F. (1973). Robust modeling with
erratic data.
Duda, R., Hart, P., and Stork, D. (2012). Pattern Classifica-
tion. Wiley-interscience, USA, 2nd edition.
Fisher, R. (1936). The use of multiple measurements in
taxonomic problems. Annals of Eugenics, 7(2).
Fukunaga, K. (1990). Introduction to Statistical Pattern
Recognition. Academic Press, USA, 2nd edition.
Haykin, S. (2009). Neural networks and learning machines.
Prentice Hall, USA, 3rd edition.
J. A. Scales, A. Gersztenkorn, S. T. and Lines, L. R. (1988).
Robust optimization methods in geophysical inverse
theory.
Ji, J. (2006). Cgg method for robust inversion and its appli-
cation to velocity-stack inversion.
Koren, Y. and Carmel, L. (2008). Robust linear dimension-
ality reduction. IEEE Transactions on Visualization
and Computer Graphics, 10(4):459–470.
Kwak, N. (2008). Principal component analysis based on
l-1 norm maximization. IEEE Trans. Pattern Analysis
and Machine Intelligence, 30(9):1672–1680.
Kwak, N. and Choi, C.-H. (2003). Feature extraction based
on ica for binary classification problems. IEEE Trans.
on Knowledge and Data Engineering, 15(6):1374–
1388.
Kwon, O.-W. and Lee, T.-W. (2004). Phoneme recognition
using ica-based feature extraction and transformation.
Signal Processing, 84(6).
Li, X., Hu, W., Wang, H., and Zhang, Z. (2010). Linear dis-
criminant analysis using rotational invariant l1 norm.
Neurocomputing, 73.
Malalur, S. S. and Manry, M. T. (2010). Multiple optimal
learning factors for feed-forward networks.
Oh, J. and Kwak, N. (2013). Generalization of linear dis-
criminant analysis using lp-norm. Pattern Recognition
Letters, 34(6):679–685.
Olavi, N. (1993). Convergence of iterations for linear equa-
tions. Birkhauser, USA, 3rd edition.
Sugiyama, M. (2007). Dimensionality reduction of mul-
timodal labeled data by fisher discriminant analysis.
Journal of Machine Learning Research, 8:1027–1061.
Theodoridis, S. and Koutroumbas, K. (2009). Pattern
Recognition. Academic Press, USA, 4th edition.
Turk, M. and Pentland, A. Face recognition using eigen-
faces.
ICPRAM2014-InternationalConferenceonPatternRecognitionApplicationsandMethods
212