热烈庆祝祖国母亲70周年！
热烈庆祝中华人民共和国70周年！Celebrate the 70th anniversary of People’s Republic of China!我爱我的祖国！I LOVE CHINA！感谢先辈，致以最崇高的敬意！Highest respect for the forerunners！
Delta Embedding Learning (ACL19) Notes
Delta Embedding Learning, Zhang et al, ACL19https://aclweb.org/anthology/papers/P/P19/P19-1322/
ACL的一篇short，方法简单地令人震惊，但是从结果看特别有效，我觉得就很离谱，就，分享一下。
就简要说说
...
Effective Adversarial Regulariztion for Machine Translation (Notes)
最近想看看别的领域的文章，博客就随意记录一下。文章是 Effective Adversarial Regulariztion for Machine Translation, Motoki Sato et al, ACL 18. https://www.aclweb.org/anthology/P1
...
Some Tensor Arithmetic
Recommend 2 websites for some tensor arithmetics for beginner (ZH), including tensor k-mode product with vector and matrix, tensor matricization …Very
...
RNN-properties
RNN computational properties:adopted from NAACL-HLT 2018 paperhttps://www.aclweb.org/anthology/N18-1205 [Recurrent Neural Networks as Weighted Languag
...
Uncovering divergent linguistic Information in word embeddings - Paper Notes
https://arxiv.org/pdf/1809.02094.pdfPaper from arxiv, Eneko Agirre, IXA NLP Group
This paper answered some very interesting questions, In word-embeddi
...
NeurlPS 2018 Word Embedding (II), FRAGE: frequency-agnostic word representation
说一下 NeurlPS 2018的第二篇Embedding相关文章。Frequency-Agnostic Word Representation https://arxiv.org/abs/1809.06858
首先作者发现了现有的word embedding的一个现象。不常见的词（low-freq
...
Python Loading Large Dataset
今天说些写代码时碰到的问题，关于如何save&load一个大的数据object。我们对比两种工具，joblib和pickle。
首先说结论,如果是numpy array，joblib各项性能效果比pickle更好，如果是原生自带的Python object，如列表套列表，pickle远远快于
...
Pytorch中tensor的批量赋值以及选择
今天来谈一谈pytorch中快速批量的赋值和选择的操作,这种方法非常简洁,而且比for循环便利i,j,k index的效率高上很多很多倍,并且可以利用gpu加速.用深度学习框架的时候,务必避免for循环出现.
我们直接上代码:
首先我们随机生成一个 2x3x4的tensor,结果:123456789
...
On the Dimensionality of Word Embedding - NeuralPS2018 简要note
今天简单说一下关于word embedding的一篇论文的大意, https://arxiv.org/abs/1812.04224 ,‘On the Dimensionality of Word Embedding’,这是一篇被NeurIPS 2018接受的Oral Paper。文章说了一件核心的事
...