Fisher Kernel Learning


Fisher kernel learning (FKL) is a technique that can be used to train a hidden Markov model or Markov random field in such a way that the trained model can be used to produce “good” Fisher kernel features. The technique is described in more detail in the following paper:

L.J.P. van der Maaten. Learning Discriminative Fisher Kernels. In Proceedings of the International Conference on Machine Learning (ICML), pages 217-224, 2011. [ PDF ]


Code provided by Laurens van der Maaten, 2011. The author of this code do not take any responsibility for damage that is the result from bugs in the provided code. This code can be used for non-commercial purposes only. Please contact the author if you would like to use this code commercially.


The code provides implementations for standard Bayes classifiers, standard Fisher kernels, and Fisher kernel learning. As a base model, Hidden Markov Models and pairwise Markov Random Fields may be used. The provided code also allows for training and evaluating classifiers such as logistic regressors, support vector machines, and large-margin nearest neighbor classifiers on the extracted features. In addition, code is provided that shows how to use the techniques on (1) a data set of online handwritten characters, (2) a data set of spoken Arabic digits, and (3) a data set of molecules that are either mutagenous or non-mutagenous. The demonstration code can also be used to reproduce the results presented in the paper.

The code and all required data sets and toolboxes are available for download here (49 MB).

Problems / Bugs / Questions?

Feel free to drop me a line.