人工智能培训

搜索

论文代码开源:分层昼夜模型的变点检测(Change-Point Detection on Hierarchical Circadian Models)

[复制链接]
admin 发表于 2018-9-15 10:39:05 | 显示全部楼层 |阅读模式
admin 2018-9-15 10:39:05 1370 0 显示全部楼层
人工智能论文代码开源:分层昼夜模型的变点检测(Change-Point Detection on Hierarchical Circadian Models)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch。我们解决了通过深度生成模型学习的潜在表现的无监督解开的问题。与目前使用证据下限(ELBO)的方法相比,我们认为VAE潜在空间中的统计独立性可以用原始的分层贝叶斯方式强制执行。为此,我们在thelatent代码的协方差矩阵上使用逆Wishart(IW)先验来增加标准VAE。通过调整IW参数,我们能够鼓励(ordiscourage)学习潜在维度的独立性。在一系列数据集(2DShapes,3DChairs,3DFaces和CelebA)上的广泛实验结果表明我们的方法优于$ \ beta $ -VAE,并且与最先进的FactorVAE竞争。我们的方法在新数据集(CorrelatedEllipses)上实现了明显更好的纠结和重建,其引入了变异因素之间的相关性。
We address the problem of unsupervised disentanglement of latentrepresentations learnt via deep generative models.In contrast to currentapproaches that operate on the evidence lower bound (ELBO), we argue thatstatistical independence in the latent space of VAEs can be enforced in aprincipled hierarchical Bayesian manner.To this effect, we augment thestandard VAE with an inverse-Wishart (IW) prior on the covariance matrix of thelatent code.By tuning the IW parameters, we are able to encourage (ordiscourage) independence in the learnt latent dimensions.Extensiveexperimental results on a range of datasets (2DShapes, 3DChairs, 3DFaces andCelebA) show our approach to outperform the $\beta$-VAE and is competitive withthe state-of-the-art FactorVAE.Our approach achieves significantly betterdisentanglement and reconstruction on a new dataset (CorrelatedEllipses) whichintroduces correlations between the factors of variation.论文代码开源:分层昼夜模型的变点检测(Change-Point Detection on Hierarchical Circadian Models) q5GAToyMorSSYuSr.jpg
URL地址:https://arxiv.org/abs/1809.04497v1     ----pdf下载地址:https://arxiv.org/pdf/1809.04497v1    ----         ----github下载地址:https://github.com/crslab/CHyVAE    ----    论文代码开源:分层昼夜模型的变点检测(Change-Point Detection on Hierarchical Circadian Models)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch,keras,至于具体是哪一个没有完全测试。
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

admin当前离线
管理员

查看:1370 | 回复:0

快速回复 返回顶部 返回列表