My Notebook

半句再见

Diffusion Model

解读最近大火的Diffusion Model

Overview 这篇博客介绍了扩散模型（Diffusion Model）的整体框架，其中包含了详细的数学推导，希望能够让读者对扩散模型有更加深刻的认识。 不同生成模型间的关联和区别 GAN：生成式对抗网络，使用一个判别器来分类真实样本和生成的样本，另外用一个生成器从噪声出发生成伪样本，两者对抗训练直到判别器无法分辨出真实样本和生成的样本。 VA...

Posted by dhx20150812 on September 24, 2022

论文笔记

Debiasing Distantly Supervised Named Entity Recognition via Causal Intervention

原文地址：https://aclanthology.org/2021.acl-long.371/ 代码地址：https://github.com/zwkatgithub/DSCAU 摘要 这篇文章使用因果干预（Causal Intervention）的方法来去除命名实体识别（NER）任务中的字典偏差的问题。作者通过SCM解释了字典偏差中的两部分偏差的来源，然后分别通过后门调...

Posted by dhx20150812 on March 18, 2022

论文笔记

Counterfactual VQA A Cause Effect Look at Language Bias

代码已开源：https://github.com/yuleiniu/cfvqa/ 背景与动机     VQA（视觉问答）系统以一张图片和一个关于这张图片形式自由、开放式的自然语言问题作为输入，以生成一条自然语言答案作为输出。它是视觉对话、视觉阅读理解和多模态推理等任务的基础。     A VQA system takes as input an image and a free...

Posted by dhx20150812 on January 18, 2022

论文笔记

Towards Controlled and Diverse Generation of Article Comments

Towards Controlled and Diverse Generation of Article Comments Introduction 自动文章评论的任务需要机器理解文章，并生成流畅的评论。这个任务有不错的应用价值，同时也有很好的研究价值。例如，不同于机器翻译或者文本摘要，自动文章评论生成可以是多样化的。对于同样一篇文章，可以从不同的角度生成不同的合适的评论。 这个任务目...

Posted by dhx20150812 on July 28, 2021

论文笔记

Towards Robust Neural Retrieval Models with Synthetic Pre-Training

Towards Robust Neural Retrieval Models with Synthetic Pre-Training 引言 传统的稀疏检索模型，如BM25和TF-IDF依赖于简单的字词匹配，而深度检索模型，如DPR，将查询句和文档分别编码为连续向量表示，然后计算其向量表示之间的相似性。 诚然，深度检索模型取得了巨大的成功，但是他们都是遵循着标准的有监督学习的方式——训练...

Posted by dhx20150812 on July 26, 2021

论文笔记

DYPLOC - Dynamic Planning of Content Using Mixed Language Models for Text Generation

DYPLOC - Dynamic Planning of Content Using Mixed Language Models for Text Generation 来自ACL21，代码已开源 Github地址 Motivation 研究的问题：观点长文本生成 目前面临的挑战： 现有的神经网络生成方法缺乏一致性，因此需要使用content planning的方法 ...

Posted by dhx20150812 on July 7, 2021

论文笔记

SetRank - Learning a Permutation-Invariant Ranking Model for Information Retrieval

SetRank: Learning a Permutation-Invariant Ranking Model for Information Retrieval 来自SIGIR20，代码已开源https://github.com/pl8787/SetRank Introduction 传统的排序学习模型通常是基于概率排序原则Probability Ranking Princi...

Posted by dhx20150812 on January 13, 2021

论文笔记

Learning Groupwise Multivariate Scoring Functions Using Deep Neural Networks

Learning Groupwise Multivariate Scoring Functions Using Deep Neural Networks 来自ICTIR2019，https://doi.org/10.1145/3341981.3344218 代码见https://github.com/tensorflow/ranking Motivation 不同于分类或...

Posted by dhx20150812 on January 12, 2021

论文笔记

Learning a Deep Listwise Context Model for Ranking Refinement

Learning a Deep Listwise Context Model for Ranking Refinement 论文来自SIGIR18，https://arxiv.org/abs/1804.05936 代码已开源：https://github.com/QingyaoAi/Deep-Listwise-Context-Model-for-Ranking-Refineme...

Posted by dhx20150812 on January 11, 2021

论文笔记

Fixed That for You - Generating Contrastive Claims with Semantic Edits

Fixed That for You: Generating Contrastive Claims with Semantic Edits 来自NAACL 2019, https://www.aclweb.org/anthology/N19-1174.pdf 代码已开源，https://github.com/chridey/fixedthat Motivation 辩论的...

Posted by dhx20150812 on January 5, 2021

论文笔记

Sentence Level Content Planning and Style Specification for Neural Text Generation

Sentence-Level Content Planning and Style Specification for Neural Text Generation 来自EMNLP 2019，https://www.aclweb.org/anthology/D19-1055.pdf 代码已开源：http://xinyuhua.github.io/Resources/emnlp1...

Posted by dhx20150812 on November 18, 2020

论文笔记

PAIR Planning and Iterative Refinement in Pre-trained Transformers for Long Text Generation

PAIR: Planning and Iterative Refinement in Pre-trained Transformers for Long Text Generation 来自EMNLP 2020：https://arxiv.org/abs/2010.02301 代码已开源：https://github.com/XinyuHua/pair-emnlp2020 ...

Posted by dhx20150812 on October 21, 2020

论文笔记

CoCon A Self-Supervised Approach for Controlled Text Generation

CoCon: A Self-Supervised Approach for Controlled Text Generation Motivation 在使用语言模型做文本生成的任务时，给定提示文字 $x_{:t-1}=\{x_1,\cdots,x_{t-1}\}$，后续的文本 $\{x_{t}, \cdots, x_{l}\}$ 是通过自回归的方式生成的： \[p\left(x_{t...

Posted by dhx20150812 on August 30, 2020

论文笔记

Content preserving text generation with attribute controls

Content preserving text generation with attribute controls 来自NIPS 2018 https://papers.nips.cc/paper/7757-content-preserving-text-generation-with-attribute-controls.pdf 代码已开源，第三方实现见 https://g...

Posted by dhx20150812 on August 16, 2020

论文笔记

Improved Semantic Representations From Tree Structured Long Short Term Memory Networks

Improved Semantic Representations From Tree-Structured Long Short-Term Memory Networks 来自ACL 2015 代码已开源，https://github.com/stanfordnlp/treelstm 第三方TensorFlow实现：https://github.com/tensorfl...

Posted by dhx20150812 on June 11, 2020

论文笔记

Reinforced Rewards Framework for Text Style Transfer

Reinforced Rewards Framework for Text Style Transfer 来自ECIR 2020 Introduction 文本风格迁移是在保留文本的核心内容的前提下，将给定文本的风格迁移为另一种目标风格的一个任务。现有的方法在训练时主要使用了word-level的目标函数，而它与任务所期望的指标（内容保留和迁移强度）不一致。任务期望的指标通常在...

Posted by dhx20150812 on May 25, 2020

论文笔记

Plug and Play Language Models A Simple Approach to Controlled Text Generation

Plug and Play Language Models: A Simple Approach to Controlled Text Generation 来自 ICLR 2020，https://arxiv.org/pdf/1912.02164.pdf 代码已开源，https://github.com/uber-research/PPLM 官方博客见 https://...

Posted by dhx20150812 on February 23, 2020

论文笔记

Toward Controlled Generation of Text

Toward Controlled Generation of Text 来自ICML 2017，https://arxiv.org/pdf/1703.00955.pdf 代码已开源：https://github.com/asyml/texar/tree/master/examples/text_style_transfer Introduction 近些年来，研究人员对...

Posted by dhx20150812 on February 11, 2020

论文笔记

SGM Sequence Generation Model for Multi-Label Classification

SGM: Sequence Generation Model for Multi-Label Classification 来自COLING 2018，https://www.aclweb.org/anthology/C18-1330.pdf 代码已开源，https://github.com/lancopku/SGM Motivation 作者认为现有的模型通常存在两点不...

Posted by dhx20150812 on January 12, 2020

论文笔记

Label Specific Document Representation for Multi-Label Text Classification

Label-Specific Document Representation for Multi-Label Text Classification 来自EMNLP 2019，https://www.aclweb.org/anthology/D19-1044/ 代码已开源： https://github.com/EMNLP2019LSAN/LSAN Introduction...

Posted by dhx20150812 on December 18, 2019

论文笔记

LexicalAT Lexical-Based Adversarial Reinforcement Training for Robust Sentiment Classification

LexicalAT: Lexical-Based Adversarial Reinforcement Training for Robust Sentiment Classification 来自EMNLP 2019，https://www.aclweb.org/anthology/D19-1554.pdf 代码已开源 https://github.com/lancopku/Le...

Posted by dhx20150812 on December 17, 2019

论文笔记

Improving Multi-turn Dialogue Modelling with Utterance ReWriter

Improving Multi-turn Dialogue Modelling with Utterance ReWriter 来自ACL 2019，https://arxiv.org/pdf/1906.07004.pdf Introduction 单轮对话建模已经取得令人瞩目的进展，但是多轮对话的表现还远远不能令人满意。作者认为，主要的原因是我们的日常对话中存在着大量的指代和...

Posted by dhx20150812 on December 16, 2019

论文笔记

Seq2Emo for Multi-label Emotion Classification Based on Latent Variable Chains Transformation

Seq2Emo for Multi-label Emotion Classification Based on Latent Variable Chains Transformation 来自AAAI 2020 https://arxiv.org/pdf/1911.02147.pdf Motivation： 当前很多方法在做文本的情绪识别任务时，都将其建模为一个多标签分类(Mu...

Posted by dhx20150812 on December 12, 2019

论文笔记

Is Word Segmentation Necessary for Deep Learning of Chinese Representations？

Is Word Segmentation Necessary for Deep Learning of Chinese Representations? 来自ACL 2019，https://www.aclweb.org/anthology/P19-1314.pdf Introduction 在中文自然语言处理领域，分词已经被认为是一个基本的预处理手段。但是作者对分词的必要性提...

Posted by dhx20150812 on December 12, 2019

论文笔记

EDA Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks

EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks 来自EMNLP 2019 short paper，https://arxiv.org/abs/1901.11196.pdf 代码已开源https://github.com/jasonwei20/...

Posted by dhx20150812 on December 12, 2019

论文笔记

A Deep Reinforced Sequence-to-Set Model for Multi-Label Text Classification

A Deep Reinforced Sequence-to-Set Model for Multi-Label Text Classification 来自AAAI 2019，https://arxiv.org/pdf/1809.03118.pdf 代码已开源 https://github.com/lancopku/Seq2Set Introduction 多标签文本分类...

Posted by dhx20150812 on December 12, 2019

LeetCode递归题解

递归专题

938. Range Sum of BST 题目链接见 https://leetcode.com/problems/range-sum-of-bst/ Given the root node of a binary search tree, return the sum of values of all nodes with value between L and R (incl...

Posted by dhx20150812 on November 15, 2019

Leetcode动态规划题解

动态规划

动态规划(Dynamic Programming) 3. Longest Substring Without Repeating Characters 题目链接 https://leetcode.com/problems/longest-substring-without-repeating-characters/ Given a string, find the length...

Posted by dhx20150812 on November 14, 2019

论文笔记

Graph Convolutional Networks for Text Classification

Graph Convolutional Networks for Text Classification Introduction 文本分类是自然语言处理中一个很基础的问题。文本分类中的一个基本的中间步骤是文本表示，传统的方法使用人工提取的特征来表示文本，如词袋和n-gram等。深度学习的模型，如CNN，LSTM等考虑到了局部性和顺序性，他们可以很好地捕获局部连续的单词中的语义和语法信息...

Posted by dhx20150812 on July 25, 2019

论文笔记

BERT论文笔记

Bert论文笔记 本文是个人在阅读Bert原论文的笔记，作为日后回顾时参考。 Motivation 预训练的语言模型已被证明在许多自然语言处理任务上很有效，包括自然语言推断，命名实体识别，智能问答等。目前有两种主流的策略将预训练的语言表示应用到下游任务上，一种策略是基于特征(feature-based)的，另一种是微调(fine-tuning)。前者的典型代表是ELMo，它使用...

Posted by dhx20150812 on April 13, 2019

FEATURED TAGS
NLP 文本分类 GCN Leetcode DP Recursive 多标签分类 数据增强 分词技术 对话系统 情感分类 对抗攻击 文本生成 可控文本生成 文本风格迁移 强化学习 Argument Generation Content Planning Contrastive Claim Generation Information Retrieval Learning to Rank Text Generation Dense Retrieval 视觉问答 反事实 因果推断 命名实体识别 因果干预 扩散模型 Diffusion Model
ABOUT ME

Accept and then overcome

✉️ mlhxding@gmail.com