Conclusion

In this wiki we have covered kernel methods. We explained how kernels can be swapped in anywhere dot products appear in a machine learning algorithm, to create a nonlinear method. The same derivation can be applied to other methods. For example, there also is a kernel PCA - although kernel PCA is not very practically useful, it provides a useful way to think about other nonlinear dimensionality reduction methods, as they can all be shown to be special cases of Kernel PCA.