arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2403.03069cs.LGstat.ML

利用混合变分族改进不完整数据下变分自编码器的估计

Improving Variational Autoencoder Estimation from Incomplete Data with Mixture Variational Families

University of Edinburgh(爱丁堡大学);School of Informatics(信息学院)

Vaidotas Simkus, Michael U. Gutmann

AI总结:

针对不完整数据下VAE估计中后验复杂度增加的问题,提出基于有限变分混合和插补变分混合的两种策略,并通过实验证明其能有效提升估计准确性。

摘要:

我们考虑在训练数据不完整的情况下估计变分自编码器(VAEs)的任务。我们表明,与完全观测的情况相比,缺失数据增加了模型在潜变量上后验分布的复杂度。由于变分后验分布与模型后验分布之间的不匹配,这种增加的复杂度可能会对模型的拟合产生不利影响。我们引入两种策略,分别基于(i)有限变分混合分布和(ii)基于插补的变分混合分布,以应对增加的后验复杂度。通过对所提方法的全面评估,我们表明变分混合分布能有效提高从不完整数据中估计VAE的准确性。

英文摘要:

We consider the task of estimating variational autoencoders (VAEs) when the training data is incomplete. We show that missing data increases the complexity of the model's posterior distribution over the latent variables compared to the fully-observed case. The increased complexity may adversely affect the fit of the model due to a mismatch between the variational and model posterior distributions. We introduce two strategies based on (i) finite variational-mixture and (ii) imputation-based variational-mixture distributions to address the increased posterior complexity. Through a comprehensive evaluation of the proposed approaches, we show that variational mixtures are effective at improving the accuracy of VAE estimation from incomplete data.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑