对数线性模型中超越成对势的凸结构学习Convex structure learning in log-linear models beyond pairwise potentials |
|
课程网址: | http://videolectures.net/aistats2010_schmidt_cslil/ |
主讲教师: | Mark Schmidt |
开课单位: | 不列颠哥伦比亚大学 |
开课时间: | 信息不详。欢迎您在右侧留言补充。 |
课程语种: | 英语 |
中文简介: | 以前的工作已经研究了l1正则化对数线性模型的结构学习,主要集中在成对势的情况下。在本工作中,我们考虑具有任意阶势的模型,但它满足一个层次约束。我们使用重叠群的组l1正则化来加强分层约束,而一种强制分层包含的主动集方法允许我们可追溯地考虑高阶电势的指数数量。摘要利用光谱投影梯度法作为求解重叠群l1正则化问题的子例程,利用稀疏的Dykstra算法计算投影。实验表明,该模型与以前的模型相比,给出了相同或更好的测试集似然。 |
课程简介: | Previous work has examined structure learning in log-linear models with L1-regularization, largely focusing on the case of pairwise potentials. In this work we consider the case of models with potentials of arbitrary order, but that satisfy a hierarchical constraint. We enforce the hierarchical constraint using group L1-regularization with overlapping groups, and an active set method that enforces hierarchical inclusion allows us to tractably consider the exponential number of higher-order potentials. We use a spectral projected gradient method as a sub-routine for solving the overlapping group L1-regularization problem, and make use of a sparse version of Dykstra's algorithm to compute the projection. Our experiments indicate that this model gives equal or better test set likelihood compared to previous models. |
关 键 词: | 对数线性模型; 凸结构 |
课程来源: | 视频讲座网 |
最后编审: | 2019-10-30:cwx |
阅读次数: | 15 |