https://www.youtube.com/watch?v=iZTeva0WSTQ&list=PLoROMvodv4rMiGQp3WXShtMGgzqpfVfbU&index=4

Perceptron - An historic machine learning algorithm which is not used much in practice.

It is not used in practice because it doesn’t have the probabilistic approach.

Exponential Family

PDF = Probability Density Function

y - Data

n - natural parameter

T(y) - sufficient statistics

b(y) - Base measure

a(n) - log partition

Bernoulli Distribution - Binary Data :

Now consider these parameters:

Then the Bernoulli distribution can be fitted in Exponential Family.

Similarly, many other distributions can be placed in this family.

Properties of Exponential Families