@InProceedings{TsuKawRaeSonMue02, author = {Koji Tsuda and Motoaki Kawanabe and Gunnar R{\"a}tsch and S\"oren Sonnenburg and Klaus-Robert M{\"u}ller}, title = {A New Discriminative Kernel from Probabilistic Models}, editor = {T.G.~Dietterich and S.~Becker and Z.~Ghahramani}, pages = {977--984}, year = {2002}, volume = {14}, booktitle = {Advances in Neural information processings systems}, publisher = {MIT Press}, address = {Cambridge, MA}, ps = {http://sonnenburgs.de/soeren/publications/TsuKawRaeSonMue02.ps.gz}, pdf = {http://sonnenburgs.de/soeren/publications/TsuKawRaeSonMue02.pdf.gz}, abstract = {Recently, Jaakkola and Haussler proposed a method for constructing ker- nel functions from probabilistic models. Their so called “Fisher kernel” has been combined with discriminative classifiers such as SVM and applied successfully in e.g. DNA and protein analysis. Whereas the Fisher kernel (FK) is calculated from the marginal log-likelihood, we propose the TOP kernel derived from Tangent vectors Of Posterior log-odds. Furthermore we develop a theoretical framework on feature extractors from probabilistic models and use it for analyzing FK and TOP. In experiments our new discriminative TOP kernel compares favorably to the Fisher kernel.