一种新颖变分自编码器及其在生成建模、分类与有序回归中的应用
机器学习
2018-12-20 v2 机器学习
摘要
我们基于变分自编码器方法开发了一种新颖的概率生成模型。我们架构的显著特点在于：一种指定潜变量先验的新颖方式，以及引入了一个保序单元。我们描述了如何利用该模型进行监督、无监督与半监督学习，以及名义分类与有序分类。我们使用两个基准数据集分析了该方法的生成特性，以及在名义与有序分类下的分类有效性。我们的结果表明，我们的模型在两项分类任务中都能与相关基线取得可比的结果。
引用
@article{arxiv.1812.07352,
title = {A Novel Variational Autoencoder with Applications to Generative Modelling, Classification, and Ordinal Regression},
author = {Joel Jaskari and Jyri J. Kivinen},
journal= {arXiv preprint arXiv:1812.07352},
year = {2018}
}
备注
The first version [v1] contains our paper submitted (on 9 February, 2018) to and later rejected from the Thirty-Fifth International Conference on Machine Learning (ICML 2018); earlier version of the paper was submitted (on 13 October, 2017 [UTC]) to and later rejected from the Twenty-First International Conference on Artificial Intelligence and Statistics (AISTATS 2018)