A Robust Regularization Path Algorithm for $\nu $ -Support Vector Classification
Citations
2,140 citations
730 citations
Additional excerpts
...org) extra transmission energy cost [7]....
[...]
365 citations
324 citations
253 citations
Cites methods from "A Robust Regularization Path Algori..."
...To demonstrate the effectiveness of the MIMAGA-Selection algorithm, we apply three existing feature selection algorithms: ReliefF [39,40] , sequential forward selection (SFS) [41,42] and MIM on the same datasets with the same target gene numbers....
[...]
References
26,531 citations
"A Robust Regularization Path Algori..." refers methods in this paper
...…first give the linear relationship between α, g, and ν (see line 4 in Algorithm 1 and Section III-A1), then compute the maximal adjustment quantity νmax (see line 5 in Algorithm 1 and Section III-A2), and finally update α, g̃, d1, d2, SS , SR , and SE (see line 6 in Algorithm 1 and Section III-A3)....
[...]
2,737 citations
"A Robust Regularization Path Algori..." refers methods in this paper
...According to convex optimization theory [13, p. 229], the solution of the dual problem with two equality constraints can be obtained by minimizing the following convex function: min 0≤αi≤1/ l Wν = 1 2 αT Qα + b′ ⎛ ⎝ l ∑ i=1 yiαi ⎞ ⎠+ ρ′ ⎛ ⎝ l ∑ i=1 αi − ν ⎞ ⎠ (3) where both b′ and ρ′ are Lagrangian multipliers....
[...]
...In order to propose the ν-SvcRPath, i.e., Algorithm 1, we first give the linear relationship between α, g, and ν (see line 4 in Algorithm 1 and Section III-A1), then compute the maximal adjustment quantity νmax (see line 5 in Algorithm 1 and Section III-A2), and finally update α, g̃, d1, d2, SS ,…...
[...]
...For each adjustment of ν (i.e., ∑ i∈S αi ), in order to keep all the samples satisfying the KKT conditions, the weights of the samples in SS , and the Lagrange multipliers (b ′ and ρ′) should also be adjusted accordingly....
[...]
...(13) According to the convex optimization theory [13], the solution of (13) can also be obtained by minimizing the following convex function: min 0≤αi≤1/ l W ′ν = 1 2 αT Qα + d1 ⎛ ⎝ ∑ i∈S+ αi − ν2 ⎞ ⎠ + d2 ⎛ ⎝ ∑ i∈S− αi − ν2 ⎞ ⎠ (14) where both d1 and d2 are Lagrangian multipliers....
[...]
...8: end while 1) Compute Linear Relationship Between α, g, and ν: For each adjustment of ν (i.e., ∑ i∈S αi ), in order to keep all the samples satisfying the KKT conditions, the weights of the samples in SS , and the Lagrange multipliers (d1 and d2) should also be adjusted accordingly....
[...]
2,140 citations
940 citations