1.
Suppose the fraction of rural population of India who owns a car is 15% and the fraction of
urban population of India who owns a car is 23%. If 1/5th of the Indian population is urban
and the rest are rural, what is the probability that a person who owns a car is an urban person?
1
2. Lagrangian function of a primal problem is given by Lp = x' Kx + c' x + ' ( Ax + d ) where K
2
is a symmetric matrix and α is Lagrangian multiplier vector. Obtain the Lagrangian function of the
dual problem.
3. A transformation ( X ) = [1, x12 , 2 x1 x2 , x22 , 2 x1 , 2 x2 ]T is used to transform the 2-dimensional
data into higher dimension. Determine the kernel function.
4. Consider a support vector machine whose input space is ℝ2, and in which the kernel function
is computed as, 𝑘(𝒙,𝒚)= (𝒙⋅𝒚+1)2−1, (bold letters represents vectors in ℝ2). Find the
mapping ɸ (𝒙) to the feature space corresponding to this kernel. Show your derivation.
5. Let X1 = (1,−1,−1), y1 =−1, X2 = (−3,1,1), y2 = 1, X3=(−3,1,−1), y3=−1, X4= (1,2,1),
y4 =−1, and X5 = (−1,−1,2). y5 = 1, be five binary labeled training examples. These points
are linearly separable. Derive the optimum margin classifier (support vectors, weights and
threshold value) and the margin.
Hint:
For this dataset, visual inspection (or solving the dual optimization problem) reveals
that the SVs are:
• X₂ = (-3, 1, 1) (label = 1)
• X₅ = (-1, -1, 2) (label = 1)
• X₄ = (1, 2, 1) (label = -1)
These points lie on the margin boundaries and satisfy yi(wTXi+b)=1.
