SMO (Sequence minimum optimization)--SVM

SMO (Sequence minimum optimization)
Introduction: Coordinate Ascent method
The goals are:

Coordinate ascent algorithm:

That is, only one dimension is changed at a time, and the optimal value of the dimension is obtained.
Example diagram:

The direction of parameter convergence is parallel to the axis.

SMO algorithm:

Because one of the constraints we're dealing with is:

So it is not possible to change only one of the variables, so you need to select two variables to change (one of the variables can be obtained from the other variable).

α may oscillate next to the optimal value, but W (α) will gradually tend to be optimal and remain unchanged at most.

SMO (Sequence minimum optimization)--SVM