as far as i understand

Dual formulation is basically for the purpose of maximizing the margin and minimizing the training error. 

reference:

http://www.cs.columbia.edu/~kathy/cs4701/documents/jason_svm_tutorial.pdf

as far as i understand

Dual formulation is basically for the purpose of maximizing the margin and minimizing the training error. 

reference:

http://www.cs.columbia.edu/~kathy/cs4701/documents/jason_svm_tutorial.pdf

We use dual formulation of SVM (or you can say dual formulation is effective in following cases), mainly when-

1. Number of features are larger than number of training points.

2. Data is non-linearly separable.

3. To utilize the sparsity of data (if present)

 4. Primal formulation suffers from differentiability issues.

Hy, 

it's a probleme transformation, 

it's brings the problem resolution and maximum hyperplan feasible and more easly.