This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
assignments:assignment3 [2015/10/02 12:12] asa |
assignments:assignment3 [2015/10/07 12:04] asa |
||
---|---|---|---|
Line 7: | Line 7: | ||
Formulate a soft-margin SVM without the bias term, i.e. one where the discriminant function is equal to $\mathbf{w}^{T} \mathbf{x}$. | Formulate a soft-margin SVM without the bias term, i.e. one where the discriminant function is equal to $\mathbf{w}^{T} \mathbf{x}$. | ||
Derive the saddle point conditions, KKT conditions and the dual. | Derive the saddle point conditions, KKT conditions and the dual. | ||
- | Compare it to the standard SVM formulation. | + | Compare it to the standard SVM formulation that was derived in class. |
- | As we discussed in class, SMO-type algorithms for the dual optimize the smallest number of variables at a time, which is two variables. | + | In class we discussed SMO-type algorithms for optimizing the dual SVM. At each step SMO optimizes two variables at a time, which is the smallest number possible. |
- | Is this still the case for the formulation you have derived? | + | Is this still the case for the formulation you have derived? In other words, is two the smallest number of variables that can be optimized at a time? |
Hint: consider the difference in the constraints. | Hint: consider the difference in the constraints. | ||
Line 64: | Line 64: | ||
K_{gauss}(\mathbf{x}, \mathbf{x'}) = \exp(-\gamma || \mathbf{x} - \mathbf{x}' ||^2) | K_{gauss}(\mathbf{x}, \mathbf{x'}) = \exp(-\gamma || \mathbf{x} - \mathbf{x}' ||^2) | ||
$$ | $$ | ||
+ | and | ||
$$ | $$ | ||
- | K_{poly}(\mathbf{x}, \mathbf{x'}) = (\mathbf{x}^T \mathbf{x}' + 1) ^{p} | + | K_{poly}(\mathbf{x}, \mathbf{x'}) = (\mathbf{x}^T \mathbf{x}' + 1) ^{p}. |
$$ | $$ | ||
Line 110: | Line 111: | ||
Part 1: 40 points. | Part 1: 40 points. | ||
(10 points): Primal SVM formulation is correct | (10 points): Primal SVM formulation is correct | ||
- | (10 points): Lagrangian found correctly | + | ( 7 points): Lagrangian found correctly |
- | (10 points): Derivation of saddle point equations | + | ( 8 points): Derivation of saddle point equations |
(10 points): Derivation of the dual | (10 points): Derivation of the dual | ||
( 5 points): Discussion of the implication of the form of the dual for SMO-like algorithms | ( 5 points): Discussion of the implication of the form of the dual for SMO-like algorithms |