稀疏算法不稳定:一个没有免费午餐的定理Sparse Algorithms are Not Stable: A No-free-lunch Theorem |
|
课程网址: | http://videolectures.net/nipsworkshops2012_xu_theorem/ |
主讲教师: | Huan Xu |
开课单位: | 新加坡国立大学 |
开课时间: | 2013-01-16 |
课程语种: | 英语 |
中文简介: | 我们考虑机器学习中两个广泛使用的概念,即:稀疏性和稳定性。这两种观念都被认为是可取的,并且被认为可以带来良好的泛化能力。我们证明这两个概念相互矛盾:稀疏算法不能稳定,反之亦然。因此,必须在设计学习算法时权衡稀疏性和稳定性。这意味着,与\ ell_2正则化回归相比,\ ell_1正则化回归(Lasso)不能稳定。 |
课程简介: | We consider two widely used notions in machine learning, namely: sparsity and stability. Both notions are deemed desirable, and are believed to lead to good generalization ability. We show that these two notions contradict each other: a sparse algorithm can not be stable and vice versa. Thus, one has to tradeoff sparsity and stability in designing a learning algorithm. This implies that, in contrast to \ell_2 regularized regression, \ell_1 regularized regression (Lasso) cannot be stable. |
关 键 词: | 机器学习; 稀疏算法; 学习算法 |
课程来源: | 视频讲座网 |
最后编审: | 2019-09-08:lxf |
阅读次数: | 28 |