In a [[Feed Forward Neural Network]] (”NN”) the feature mapping $x \mapsto \phi(x)$ is not fixed, unlike in other models (e.g. [[Polynomial Kernel Function]]). We can split a NN into two components: - The first $n$ layers create a (complex) feature mapping $\phi(x)$ - The output layer acts as a [[Linear Classifier]] based on that $\phi(x)$. ![[neural-networks-feature-mapping.png|center|400]]