A type of Classification method.

Logistic regression Definition

pdf

See also here for intro

Probabilistic interpretation. Logitistic regression can also be interpretedas a Maximum likelihood estimate assuming the Logistic funciton (sigmoid) gives the probability of belonging to a given class. We can then use Gradient descent. More theoretical understanding can be acquired from the theory of Generalized linear models

A simpler version: Perceptron

See here

Why is it convex

Training

Maximum likelihood. Log likelihood turns out to be cross-entropy.

To see if problem is convex, we test if the Hessian is positive semi-definite, and it is. See here for form of Hessian

Kernel logistic regression

Video