Fused Lasso模型的特征选择

Research on Feature Selection of Fused Lasso

该文考虑了一类Fused Lasso问题的特征选择方法。与之前的方法不同,该文利用变分不等式为对偶问题提供充要条件,构造了特征选择方法。通过给出优化问题的对偶问题,进而导出对偶问题变分不等式形式下的必要条件。构造一个包含对偶最优解的对偶可行域,并在这个可行域上估计对偶约束上界,建立筛选规则,识别出具有相同系数的相邻特征,进而实现特征剔除。

This paper considers the feature selection for Fused Lass.Unlike the previous method,this paper uses variational inequality to provide sufficient and necessary conditions for the dual problem,and constructs the feature selection.By giving the dual problem of the optimization problem,derive the necessary conditions in the form of variational inequality of the dual problem.Construct a dual feasible region containing the dual optimal solution,and estimate the upper bound of the dual constraint on this feasible region.Established a screening rule,to identify adjacent features with the same coefficient,and achieve feature removal.