Distances and kernels based on cumulative distribution functions - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 474

Our distance measures can be extended to high-dimensional spaces. However, the calcula-
tion of distance using CDFs may incur heavy computational cost. For future work, we propose
to develop an algorithm that computes the distance measure directly from the data, bypassing
the explicit construction of CDFs.
References
[1] Boser BE, Guyon IM, Vapnik VN. A training algorithm for optimal margin classifiers. In:
Proc. fifth annual workshop on computational learning theory - COLT '92; 1992:144.
[2] Su H, Zhang H. Distances and kernels based on cumulative distribution functions. In: Proc.
2014 international conference on image processing, computer vision, & patern recog-
nition; 2014:357-361.
[3] Bhatacharyya A. On a measure of divergence between two statistical populations
deined by their probability distributions. Bull Calcuta Math Soc. 1943;35:99-109.
[4] Jebara T, Kondor R, Howard A. Probability product kernels. J Mach Learn Res.
2004;5:819-844.
[5] Hellinger E. Neue Begründung der Theorie quadratischer Formen von unend-
lichvielen Veränderlichen. J Reine Angew Math (in German). 1909;136:210-271.
[6] Kullback S, Leibler RA. On information and sufficiency. Ann Math Stat.
1951;22(1):79-86.
[7] Bogachev VI, Kolesnikov AV. The Monge-Kantorovich problem: achievements, con-
nections, and perspectives. Russ Math Surv. 2012;67:785-890.
[8] Smirnov NV. Approximate distribution laws for random variables, constructed from
empirical data. Uspekhi Mat Nauk. 1944;10:179-206 (In Russian).
[9] Greton A, Borgwardt KM, Rasch MJ, Scholkopf B, Smola A. A kernel two-sample
test. J Mach Learn Res. 2012;13:723-773.
[10] Munkres J. Algorithms for the assignment and transportation problems. J Soc Ind Appl
Math. 1957;5(1):32-38.
[11] Bishop C. Patern recognition and machine learning. New York: Springer; 2006.