Skip to content

Commit

Permalink
lbfgs
Browse files Browse the repository at this point in the history
  • Loading branch information
endymecy committed Jan 24, 2017
1 parent 3172d2c commit b3162e1
Showing 1 changed file with 4 additions and 4 deletions.
8 changes: 4 additions & 4 deletions 最优化算法/L-BFGS/lbfgs.md
Original file line number Diff line number Diff line change
Expand Up @@ -151,13 +151,13 @@ $$J(x) = l(x) + r(x)$$

  根据对模型参数所服从的概率分布的假设的不同,常用的正则化一般有`L2`正则化(模型参数服从`Gaussian`分布)、`L1`正则化(模型参数服从`Laplace`分布)以及它们的组合形式。

  `L1`正则化的形式如下
  `L1`正则化的形式如下

$$ J(x) = l(x) + r(x) = l(x) + C||x||_{1} = l(x) + C\sum_{i} |x_{i}| $$
$$J(x) = l(x) + r(x) = l(x) + C ||x||_{1} = l(x) + C \sum_{i} |x_{i}|$$

  `L2`正则化的形式如下
  `L2`正则化的形式如下

$$ J(x) = l(x) + r(x) = l(x) + C||x||_{2} = l(x) + C\sum_{i} x_{i}^{2} $$
$$J(x) = l(x) + r(x) = l(x) + C ||x||_{2} = l(x) + C \sum_{i} x_{i}^{2}$$

  `L1`正则化和`L2`正则化之间的一个最大区别在于前者可以产生稀疏解,这使它同时具有了特征选择的能力,此外,稀疏的特征权重更具有解释意义。如下图:

Expand Down

0 comments on commit b3162e1

Please sign in to comment.