Summary
This chapter has been focused on kernel methods, which are also called kernelized algorithms. The chapter has been short so that we can focus on the most important concepts underpinning kernel methods. Those concepts are as follows:
- Inner-product based learning algorithms are very common because an inner product captures the similarity between feature vectors, and learning by similarity is a natural basis for many machine learning algorithms.
- Inner products calculated from the existing features on a dataset may not be sufficient to learn the non-linear structure present in the dataset.
- Construction of new features can be necessary to make our learning algorithms accurate.
- Mercer’s theorem tells us that positive semi-definite kernel functions implicitly construct new features and calculate inner products in those new feature spaces.
- There are different types of kernel functions.
- We can use the kernel trick to kernelize any inner-product based...
