Kernel Fisher discriminant analysis

In statistics, kernel Fisher discriminant analysis (KFD),[1] also known as generalized discriminant analysis[2] and kernel discriminant analysis,[3] is a kernelized version of linear discriminant analysis (LDA). It is named after Ronald Fisher.

Linear discriminant analysis

edit

Intuitively, the idea of LDA is to find a projection where class separation is maximized. Given two sets of labeled data,   and  , we can calculate the mean value of each class,   and  , as

 

where   is the number of examples of class  . The goal of linear discriminant analysis is to give a large separation of the class means while also keeping the in-class variance small.[4] This is formulated as maximizing, with respect to  , the following ratio:

 

where   is the between-class covariance matrix and   is the total within-class covariance matrix:

 

The maximum of the above ratio is attained at

 

as can be shown by the Lagrange multiplier method (sketch of proof):

Maximizing   is equivalent to maximizing

 

subject to

 

This, in turn, is equivalent to maximizing  , where   is the Lagrange multiplier.

At the maximum, the derivatives of   with respect to   and   must be zero. Taking   yields

 

which is trivially satisfied by   and  

Extending LDA

edit

To extend LDA to non-linear mappings, the data, given as the   points   can be mapped to a new feature space,   via some function   In this new feature space, the function that needs to be maximized is[1]

 

where

 

and

 

Further, note that  . Explicitly computing the mappings   and then performing LDA can be computationally expensive, and in many cases intractable. For example,   may be infinite dimensional. Thus, rather than explicitly mapping the data to  , the data can be implicitly embedded by rewriting the algorithm in terms of dot products and using kernel functions in which the dot product in the new feature space is replaced by a kernel function, .

LDA can be reformulated in terms of dot products by first noting that   will have an expansion of the form[5]

 

Then note that

 

where

 

The numerator of   can then be written as:

 

Similarly, the denominator can be written as

 

with the   component of   defined as   is the identity matrix, and   the matrix with all entries equal to  . This identity can be derived by starting out with the expression for   and using the expansion of   and the definitions of   and  

 

With these equations for the numerator and denominator of  , the equation for   can be rewritten as

 

Then, differentiating and setting equal to zero gives

 

Since only the direction of  , and hence the direction of   matters, the above can be solved for   as

 

Note that in practice,   is usually singular and so a multiple of the identity is added to it [1]

 

Given the solution for  , the projection of a new data point is given by[1]

 

Multi-class KFD

edit

The extension to cases where there are more than two classes is relatively straightforward.[2][6][7] Let   be the number of classes. Then multi-class KFD involves projecting the data into a  -dimensional space using   discriminant functions

 

This can be written in matrix notation

 

where the   are the columns of  .[6] Further, the between-class covariance matrix is now

 

where   is the mean of all the data in the new feature space. The within-class covariance matrix is

 

The solution is now obtained by maximizing

 

The kernel trick can again be used and the goal of multi-class KFD becomes[7]

 

where   and

 

The   are defined as in the above section and   is defined as

 

  can then be computed by finding the   leading eigenvectors of  .[7] Furthermore, the projection of a new input,  , is given by[7]

 

where the   component of   is given by  .

Classification using KFD

edit

In both two-class and multi-class KFD, the class label of a new input can be assigned as[7]

 

where   is the projected mean for class   and   is a distance function.

Applications

edit

Kernel discriminant analysis has been used in a variety of applications. These include:

See also

edit

References

edit
  1. ^ a b c d e Mika, S; Rätsch, G.; Weston, J.; Schölkopf, B.; Müller, KR (1999). "Fisher discriminant analysis with kernels". Neural Networks for Signal Processing IX: Proceedings of the 1999 IEEE Signal Processing Society Workshop (Cat. No.98TH8468). Vol. IX. pp. 41–48. CiteSeerX 10.1.1.35.9904. doi:10.1109/NNSP.1999.788121. ISBN 978-0-7803-5673-3. S2CID 8473401.
  2. ^ a b c Baudat, G.; Anouar, F. (2000). "Generalized discriminant analysis using a kernel approach". Neural Computation. 12 (10): 2385–2404. CiteSeerX 10.1.1.412.760. doi:10.1162/089976600300014980. PMID 11032039. S2CID 7036341.
  3. ^ a b Li, Y.; Gong, S.; Liddell, H. (2003). "Recognising trajectories of facial identities using kernel discriminant analysis". Image and Vision Computing. 21 (13–14): 1077–1086. CiteSeerX 10.1.1.2.6315. doi:10.1016/j.imavis.2003.08.010.
  4. ^ Bishop, CM (2006). Pattern Recognition and Machine Learning. New York, NY: Springer.
  5. ^ Scholkopf, B; Herbrich, R.; Smola, A. (2001). "A Generalized Representer Theorem". Computational Learning Theory. Lecture Notes in Computer Science. Vol. 2111. pp. 416–426. CiteSeerX 10.1.1.42.8617. doi:10.1007/3-540-44581-1_27. ISBN 978-3-540-42343-0.
  6. ^ a b Duda, R.; Hart, P.; Stork, D. (2001). Pattern Classification. New York, NY: Wiley.
  7. ^ a b c d e Zhang, J.; Ma, K.K. (2004). "Kernel fisher discriminant for texture classification". {{cite journal}}: Cite journal requires |journal= (help)
  8. ^ Liu, Q.; Lu, H.; Ma, S. (2004). "Improving kernel Fisher discriminant analysis for face recognition". IEEE Transactions on Circuits and Systems for Video Technology. 14 (1): 42–49. doi:10.1109/tcsvt.2003.818352. S2CID 39657721.
  9. ^ Liu, Q.; Huang, R.; Lu, H.; Ma, S. (2002). "Face recognition using kernel-based Fisher discriminant analysis". IEEE International Conference on Automatic Face and Gesture Recognition.
  10. ^ Kurita, T.; Taguchi, T. (2002). "A modification of kernel-based Fisher discriminant analysis for face detection". Proceedings of Fifth IEEE International Conference on Automatic Face Gesture Recognition. pp. 300–305. CiteSeerX 10.1.1.100.3568. doi:10.1109/AFGR.2002.1004170. ISBN 978-0-7695-1602-8. S2CID 7581426.
  11. ^ Feng, Y.; Shi, P. (2004). "Face detection based on kernel fisher discriminant analysis". IEEE International Conference on Automatic Face and Gesture Recognition.
  12. ^ Yang, J.; Frangi, AF; Yang, JY; Zang, D., Jin, Z. (2005). "KPCA plus LDA: a complete kernel Fisher discriminant framework for feature extraction and recognition". IEEE Transactions on Pattern Analysis and Machine Intelligence. 27 (2): 230–244. CiteSeerX 10.1.1.330.1179. doi:10.1109/tpami.2005.33. PMID 15688560. S2CID 9771368.{{cite journal}}: CS1 maint: multiple names: authors list (link)
  13. ^ Wang, Y.; Ruan, Q. (2006). "Kernel fisher discriminant analysis for palmprint recognition". International Conference on Pattern Recognition.
  14. ^ Wei, L.; Yang, Y.; Nishikawa, R.M.; Jiang, Y. (2005). "A study on several machine-learning methods for classification of malignant and benign clustered microcalcifications". IEEE Transactions on Medical Imaging. 24 (3): 371–380. doi:10.1109/tmi.2004.842457. PMID 15754987. S2CID 36691320.
  15. ^ Malmgren, T. (1997). "An iterative nonlinear discriminant analysis program: IDA 1.0". Computer Physics Communications. 106 (3): 230–236. Bibcode:1997CoPhC.106..230M. doi:10.1016/S0010-4655(97)00100-8.
edit