证明两者等价
令 ξ i = [ 1 − y i ( W X i + b ) ] + \xi_i=[1-y_i(WX_i+b)]_+ ξi=[1−yi(WXi+b)]+ (4)则 ξ i ≥ 0 \xi_i\geq0 ξi≥0 and y i ( W T X i + b ) ≥ 1 − ξ i y_{i}\left(W^{T} X_{i}+b\right) \geq1-\xi_i yi(WTXi+b)≥1−ξi
so W,b, ξ i \xi_i ξi满足(1)(2)
设 W ∗ , b ∗ W^*,b^* W∗,b∗为(3)的最优解,则 W ∗ , b ∗ , ξ ∗ W^*,b^*,\xi^* W∗,b∗,ξ∗满足(1)(2)【 ξ \xi ξ由(4)构造】
( 0 ) ∣ W ∗ , b ∗ , ξ ∗ = ( 3 ) ∣ W ∗ , b ∗ ≤ ( 3 ) ∣ W , b = ( 0 ) ∣ W , b , ξ ( 构 造 ) ≤ ( 0 ) ∣ W , b , ξ ( 任 意 ) (0)|_{W^*, b^*, \xi^*}=(3)|_{W^*, b^*}\leq(3)|_{W, b}=(0)|_{W, b,\xi(构造)}\leq(0)|_{W, b,\xi(任意)} (0)∣W∗,b∗,ξ∗=(3)∣W∗,b∗≤(3)∣W,b=(0)∣W,b,ξ(构造)≤(0)∣W,b,ξ(任意)
=:由构造
<=:定义
=:由构造
<=:for any W , b , ξ W,b,\xi W,b,ξ satisfy (1)&(2) ,suppose some ξ i ≠ [ 1 − y i ( W X i + b ) ] + \xi_i\neq[1-y_i(WX_i+b)]_+ ξi=[1−yi(WXi+b)]+,if 1 − y i ( W X i + b ) ≤ 0 1-y_i(WX_i+b)\leq0 1−yi(WXi+b)≤0 , then ξ i > 0 \xi_i>0 ξi>0, 令 ξ i = 0 = [ 1 − y i ( W X i + b ) ] + \xi_i=0=[1-y_i(WX_i+b)]_+ ξi=0=[1−yi(WXi+b)]+可使(0)更小。if 1 − y i ( W X i + b ) > 0 1-y_i(WX_i+b)>0 1−yi(WXi+b)>0,then ξ i > 1 − y i ( W X i + b ) \xi_i>1-y_i(WX_i+b) ξi>1−yi(WXi+b) ,令 ξ i = 1 − y i ( W X i + b ) = [ 1 − y i ( W X i + b ) ] + \xi_i=1-y_i(WX_i+b)=[1-y_i(WX_i+b)]_+ ξi=1−yi(WXi+b)=[1−yi(WXi+b)]+ 可使(0)更小。故在Wb确定的情况下,最优的 ξ \xi ξ必是(4)中所构造的.
so W ∗ , b ∗ , ξ ∗ W^*,b^*,\xi^* W∗,b∗,ξ∗为(012)的最优解
设 W ∗ , b ∗ , ξ ∗ W^*,b^*,\xi^* W∗,b∗,ξ∗为(012)的最优解,则 ξ i ∗ = [ 1 − y i ( W X i + b ) ] + \xi_i^*=[1-y_i(WX_i+b)]_+ ξi∗=[1−yi(WXi+b)]+
( 3 ) ∣ W ∗ , b ∗ = ( 0 ) ∣ W ∗ , b ∗ , ξ ∗ ≤ ( 0 ) ∣ W , b , ξ ( 构 造 ) = ( 3 ) ∣ W , b (3)|_{W^*, b^*}=(0)|_{W^*, b^*, \xi^*}\leq(0)|_{W, b,\xi(构造)}=(3)|_{W, b} (3)∣W∗,b∗=(0)∣W∗,b∗,ξ∗≤(0)∣W,b,ξ(构造)=(3)∣W,b
so W ∗ , b ∗ W^*,b^* W∗,b∗为(3)的最优解