Basis Function

Published on: December 22, 2023

This is an attempt to structure my own learning and ideas on various topics, and are not intended as comprehensive educational material.

A simple regression is given by the following equation:

y(x, w) = w0 + w1 * x1 + w2 * x2 + ... + wD * xD


This function is a linear with parameters w and x. This adds significant limitations to the abilities of a regression models. To overcome this limitation we can wrap the input variable x to some non-linear fuction i.e linear combination of non-linear functions of the input variable. This allows to model complex realtionships between the input and output.


Some examples of basis functions are as follow:

Polynomial Basis Function

$$y(x,w) = \sum_{i=1}^{M-1} w_j\phi_j(x)$$

ϕj(x) is the basis functions.

Here, w = (w0,w1,⋯,wM − 1)T and ϕ = (ϕ0,ϕ1,⋯,ϕM − 1)T

Gaussian Basis Function

$$\phi_j(x) = exp\{\frac{-(x-\mu_j)^2}{2s^2}\}$$ Here, μj represents the location of the basis function. s is their spatial scale.

Sigmoidal Basis Function

$$\phi_j = \sigma(\frac{x-\mu_j}{s})$$ σ is the sigmoid function given by: $$\sigma(a) = \frac{1}{1+exp(-a)}$$

References

- Bishop, Christopher M. Pattern Recognition and Machine Learning.