Inference of Co-occurring Classes: Multi-class and Multi-label Classification - Computational Intelligence Paradigms in Advanced Pattern Classification - page 193

9 Summary
This chapter presented the main methods of multi-class and multi-label classifica-
tion. These methods can be applied to a large variety of applications and research
fields that relate to human knowledge, cognition and behavior. An example was
given of a system that infers co-occurring affective states from their non-verbal ex-
pressions in speech. Unlike other fields, the field of affective states has no definite
“ground truth” for verification of the annotation. In addition, the choice of taxono-
my has an important effect on the design. The classification method had to accom-
modate two more requirements. Firstly, different features distinguish different pairs
of classes (sparsity). Secondly, several levels of recognition were required. These
requirements are common to various cues of human behavior and knowledge, and
to a very wide variety of applications in various modalities. The classification re-
sults reflect shades of affective states and nuances of expressions and not only their
detection. They also reveal connections between complex concepts and complex
behavioral expressions, which may contribute to the understanding of human affec-
tive and social behavior and it development. This example shows that indeed such
methods can contribute to a wide range of research and applications.
Acknowledgments. The author thanks Deutsche Telekom Laboratories at Ben-Gurion
University of the Negev for their partial support of this research.
References
[1] Aggarwal, C.C.: Data streams: An overview and scientific applications. In: Gaber,
M.M. (ed.) Scientific Data Mining and Knowledge Discovery: Principles and Foun-
dations. Springer, Heidelberg (2010)
[2] Allwein, E.L., Schapire, R., Singer, Y.: Reducing multiclass to binary: a unifying ap-
proach for margin classiers. J. of Machine Learning Research 1, 113-141 (2001)
[3] Al-Naymat, G.: Data mining and discovery of astronomical knowledge. In: Gaber,
M.M. (ed.) Scientific data mining and knowledge discovery: Principles and founda-
tions. Springer, Heidelberg (2010)
[4] Amit, Y., Dekel, O., Singer, Y.: A boosting algorithm for label covering in multilabel
problems. In: Proc. AISTATS (2007)
[5] Barutcuoglu, Z., Schapire, R.E., Troyanskaya, O.G.: Hierarchical multi-label predic-
tion of gene function. Bioinformatics 22(7), 830-836 (2006)
[6] Baron-Cohen, S., Golan, O., Wheelwright, S., Hill, J.J.: Mindreading: The interactive
guide to emotions. Jessica Kingsley Limited London (2004),
http://www.jkp.com
[7] Bengio, S., Pereira, F., Singer, Y., Strelow, D.: Group Sparse Coding (2009)
[8] Boutell, M.R., Luo, J., Shen, X., Brown, C.M.: Learning multi-label scene classifica-
tion. Pattern Recognition 37(9), 1757-1771 (2004)
[9] Bradley, R.A., Terry, M.E.: Rank analysis of incomplete block designs: I. The me-
thods of paired comparisons. Biometrika 39, 324-345 (1952)
[10] Bredensteiner, E.J., Bennett, K.P.: Multicategory classification by support vector ma-
chines. Computational Optimization and Applications 12, 53-79 (1999)