人工智能培训

搜索

人工智能教程:在线折叠重要抽样的大概知识编辑(Approximate Knowledge Compilation by Online Collapsed Imp

[复制链接]
bigrc 发表于 2018-6-1 08:56:12 | 显示全部楼层 |阅读模式
bigrc 2018-6-1 08:56:12 791 0 显示全部楼层
人工智能教程:在线折叠重要抽样的大概知识编辑(Approximate Knowledge Compilation by Online Collapsed Importance  Sampling)我们介绍了折叠编译,一种新的近似推断算法的离散概率图形模型。这是一个折叠的采样算法,它根据到目前为止获得的部分样本逐步选择下一个要采样的变量。这种在线崩溃,连同对其余变量的知识汇编推断,自然地利用了分布中的地区结构和上下文相关的独立性。这些属性在确切的推理中被自然地利用,但是对于近似推断而言很困难。此外,通过在抽样期间提供部分编译电路,折叠编译可以获得针对重要性抽样的高效提案分发。我们的实验评估显示,折叠编译在标准基准上表现良好。特别是,当确切推理的数量相同时,崩溃的编译与现有技术相比具有竞争力,并且在几个基准测试中表现出优异的性能。
We introduce collapsed compilation, a novel approximate inference algorithmfor discrete probabilistic graphical models.It is a collapsed samplingalgorithm that incrementally selects which variable to sample next based on thepartial sample obtained so far.This online collapsing, together with knowledgecompilation inference on the remaining variables, naturally exploits localstructure and context- specific independence in the distribution.Theseproperties are naturally exploited in exact inference, but are difficult toharness for approximate inference.More- over, by having a partially compiledcircuit available during sampling, collapsed compilation has access to a highlyeffective proposal distribution for importance sampling.Our experimentalevaluation shows that collapsed compilation performs well on standardbenchmarks.In particular, when the amount of exact inference is equallylimited, collapsed compilation is competitive with the state of the art, andoutperforms it on several benchmarks.人工智能教程:在线折叠重要抽样的大概知识编辑(Approximate Knowledge Compilation by Online Collapsed Importance  Sampling) yaHFsBsFsz6acAfB.jpg
URL地址:https://arxiv.org/abs/1805.12565     ----pdf下载地址:http://arxiv.org/pdf/1805.12565    ----人工智能教程:在线折叠重要抽样的大概知识编辑(Approximate Knowledge Compilation by Online Collapsed Importance  Sampling)
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

bigrc当前离线
新手上路

查看:791 | 回复:0

快速回复 返回顶部 返回列表