人工智能培训

搜索

论文代码开源:时间序列分类的深度学习:综述(Deep learning for time series classification: a review)

[复制链接]
admin 发表于 2018-9-15 09:42:02 | 显示全部楼层 |阅读模式
admin 2018-9-15 09:42:02 837 0 显示全部楼层
人工智能论文代码开源:时间序列分类的深度学习:综述(Deep learning for time series classification: a review)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。Lake和Baroni(2018)最近推出了SCAN数据集,它包含与动作序列配对的简单命令,旨在测试循环序列到序列模型的强大泛化能力。他们的初始实验表明,这些模型可能会失败,因为他们缺乏提取系统规则的能力。在这里,我们仔细研究一下SCAN,并表明它并不总是捕获它所设计的泛化类型。为了缓解这个问题,我们提出了一个补充数据集,它需要将映射操作恢复为原始命令,称为NACS。我们展示了在SCAN上表现良好的模型在NACS上不一定表现良好,而且NACS表现出的性质与实际用例 - 序列 - 序列模型更紧密地对齐。
Lake and Baroni (2018) recently introduced the SCAN data set, which consistsof simple commands paired with action sequences and is intended to test thestrong generalization abilities of recurrent sequence-to-sequence models.Theirinitial experiments suggested that such models may fail because they lack theability to extract systematic rules.Here, we take a closer look at SCAN andshow that it does not always capture the kind of generalization that it wasdesigned for.To mitigate this we propose a complementary dataset, whichrequires mapping actions back to the original commands, called NACS.We showthat models that do well on SCAN do not necessarily do well on NACS, and thatNACS exhibits properties more closely aligned with realistic use-cases forsequence-to-sequence models.论文代码开源:时间序列分类的深度学习:综述(Deep learning for time series classification: a review) ZCDyYT9aDZeNulzd.jpg
URL地址:https://arxiv.org/abs/1809.04640v1     ----pdf下载地址:https://arxiv.org/pdf/1809.04640v1    ----         ----github下载地址:https://github.com/facebookresearch/NACS    ----    论文代码开源:时间序列分类的深度学习:综述(Deep learning for time series classification: a review)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:837 | 回复:0

快速回复 返回顶部 返回列表