二元概率模型的稀疏、谱特征及其它参数化
本文研究与二进制数据集上概率分布参数化有关的问题。
This paper studies issues relating to theparameterization of probability distributions over binary data sets.
二进制数据模型的几种参数化是已知的，包括Ising、广义Ising、正则化和全参数化。
Several such parameterizations of modelsfor binary data are known, including the Ising, generalized Ising, canonicaland full parameterizations.
我们还讨论了一种称之为“谱参数化”的参数化方法，它在现有文献中的介绍非常少。
We also discuss a parameterization that wecall the “spectral parameterization”, which has received significantly lesscoverage in existing literature.
我们根据正交Walsh Hadamard谐波展开设计对数线性模型，并提供了相应的参数化谱特征解释。
We provide this parameterization with aspectral interpretation by casting loglinear models in terms of orthogonalWalsh Hadamard harmonic expansions.
使用不同的标准和群稀疏正则化结构学习，我们提供了一种对于参数化全面的理论和实证比较。
Using various standard and group sparseregularizers for structural learning, we provide a comprehensive theoreticaland empirical comparison of these parameterizations.
我们证明了谱参数化与正则化具有最好的性能和稀疏度，而谱参数化不依赖于任何特定的参考状态。
We show that the spectral parameterization,along with the canonical, has the best performance and sparsity levels, whilethe spectral does not depend on any particular reference state.
与本文相关的一个英文网站供参考：
https://www.cs.ubc.ca/~schmidtm/Software/thesis.html
下载原文地址：
http://page2.dfpan.com/fs/5l6c2jf222315259166/
更多精彩文章请关注微信号：
