核心思想

基于阅读理解中QA系统的样本中可能混有对抗样本的情况,在寻找答案时,首先筛选出可能包含答案的句子,再做进一步推断。

方法

Part 1

given: 段落C   query Q

段落切分成句子:

每个句子和Q合并:

使用依存句法分析得到表示:

基于T Si T Q ,分别构建 Tree-LSTMSi  Tree-LSTMQ

两个Tree-LSTMs的叶结点的输入都是GloVe word vectors

输出隐向量分别是  hSi  hQ

hSi  hQ连接起来并传递给一个前馈神经网络来计算出Si包含Q的答案的可能性

loss 和前馈神经网络follows语义相关性网络

有监督的训练时,si包含答案为1,否则为0。

Part 2

计算最可能答案:

L代表QA似然神经网络预测的似然

将一对句子S*和Q传递给预先训练好的单BiDAF(Seo et al., 2016),生成Q的答案a^。

实验

数据集:sampled from the training set of SQuAD v1.1

there are 87,599 queries of 18,896 paragraphs in the training set of SQuAD v1.1. While each query refers to one paragraph, a paragraph may refer to multiple queries.

d=87,599 is the number of queries. The set D contains 440,135 sentence pairs, among which 87,306 are positive instances and 352,829 are negative instances.

positive instance: ,前者包含后者的答案。

两种采样方法: pair-level sampling ,paragraph-level sampling

1. In pair-level sampling, 45,000 positive instances and 45,000 negative instances are randomly selected from D as the training set.

2. paragraph-level sampling 首先随机选Qk,然后从Dk中随机采样出一个positive instance 和一个negative instance

Each set has 90,000 instances. The validation set with 3,000 instances are sampled through these two methods as well.

测试集:ADDANY adversarial dataset : 1,000 paragraphs and each paragraph refers to only one query. By splitting and combining, 6,154 sentence pairs are obtained.

实验设置:The dimension of GloVe word vectors (Pennington et al., 2014) is set as 300. The sentence scoring neural network is trained by Adagrad (Duchi et al., 2011) with a learning rate of 0.01 and a batch size of 25. Model parameters are regularized by a 10-4 strength of per-minibatch L2 regularization.

结果

评价标准:Macro-averaged F1 score (Rajpurkar et al., 2016; Jia and Liang, 2017).

对于table2,可以理解为二分类问题。

consider three types of sentences: adversarial sentences, answer sentences, and the sentences that include the answers returned by the single BiDAF system.

the x-axis denotes the ranked position for each sentence according to its likelihood score , while the y-axis is the number of sentences for each type ranked at this position.

It shows that among the 1,000 (C;Q) pairs, 647 and 657 answer sentences are selected by the QA Likelihood neural network based on pair-level sampling and paragraph-level sampling respectively, but only 136 and 141 adversarial sentences are selected by the QA Likelihood neural network.

结论

对于ADDSENT的没有做。

论文阅读 | Tackling Adversarial Examples in QA via Answer Sentence Selection的更多相关文章

  1. [论文阅读笔记] Adversarial Learning on Heterogeneous Information Networks

    [论文阅读笔记] Adversarial Learning on Heterogeneous Information Networks 本文结构 解决问题 主要贡献 算法原理 参考文献 (1) 解决问 ...

  2. [论文阅读笔记] Adversarial Mutual Information Learning for Network Embedding

    [论文阅读笔记] Adversarial Mutual Information Learning for Network Embedding 本文结构 解决问题 主要贡献 算法原理 实验结果 参考文献 ...

  3. 论文阅读 | Universal Adversarial Triggers for Attacking and Analyzing NLP

    [code] [blog] 主要思想和贡献 以前,NLP中的对抗攻击一般都是针对特定输入的,那么他们对任意的输入是否有效呢? 本文搜索通用的对抗性触发器:与输入无关的令牌序列,当连接到来自数据集的任何 ...

  4. 论文阅读 | Combating Adversarial Misspellings with Robust Word Recognition

    对抗防御可以从语义消歧这个角度来做,不同的模型,后备模型什么的,我觉得是有道理的,和解决未登录词的方式是类似的,毕竟文本方面的对抗常常是修改为UNK来发生错误的.怎么使用backgroud model ...

  5. 论文阅读 | Real-Time Adversarial Attacks

    摘要 以前的对抗攻击关注于静态输入,这些方法对流输入的目标模型并不适用.攻击者只能通过观察过去样本点在剩余样本点中添加扰动. 这篇文章提出了针对于具有流输入的机器学习模型的实时对抗攻击. 1 介绍 在 ...

  6. 论文阅读 | Generating Fluent Adversarial Examples for Natural Languages

    Generating Fluent Adversarial Examples for Natural Languages   ACL 2019 为自然语言生成流畅的对抗样本 摘要 有效地构建自然语言处 ...

  7. 《Explaining and harnessing adversarial examples》 论文学习报告

    <Explaining and harnessing adversarial examples> 论文学习报告 组员:裴建新   赖妍菱    周子玉 2020-03-27 1 背景 Sz ...

  8. 【论文阅读】Deep Adversarial Subspace Clustering

    导读: 本文为CVPR2018论文<Deep Adversarial Subspace Clustering>的阅读总结.目的是做聚类,方法是DASC=DSC(Deep Subspace ...

  9. Adversarial Examples for Semantic Segmentation and Object Detection 阅读笔记

    Adversarial Examples for Semantic Segmentation and Object Detection (语义分割和目标检测中的对抗样本) 作者:Cihang Xie, ...

随机推荐

  1. retrying failed action with response code: 403 错误解决

    [2019-06-10T06:52:51,610][INFO ][logstash.outputs.elasticsearch] retrying failed action with respons ...

  2. google chrome调试

    1,同样的代码使用firefox运行不会报错,正常运行. 2,同样的代码使用google chrome 有时候会误报网页? 如下:代码可以在 firfox edge正常运行,在chrome  oper ...

  3. sb 的长度 和 文件大小

    StringBuilder sb = new StringBuilder(); ;i<;i++)  //1 0000 0000  1亿项 { sb.AppendFormat("{0}, ...

  4. Appium Python测试环境搭建

    详细参考地址:https://www.cnblogs.com/amoyshmily/p/10500687.html 1,Appium安装:https://github.com/appium/appiu ...

  5. on(events,[selector],[data],fn) 在选择元素上绑定一个或多个事件的事件处理函数

    on(events,[selector],[data],fn) 概述 在选择元素上绑定一个或多个事件的事件处理函数.大理石平台精度等级 on()方法绑定事件处理程序到当前选定的jQuery对象中的元素 ...

  6. neo4j 一些常用的CQL

    创建节点.关系 创建节点(小明):create (n:people{name:’小明’,age:’18’,sex:’男’}) return n; 创建节点(小红): create (n:people{ ...

  7. 数据结构实验之链表六:有序链表的建立(SDUT 2121)

    #include <bits/stdc++.h> using namespace std; struct node { int data; struct node *next; }; in ...

  8. 在docker容器中python3.5环境下使用DIGITS训练caffe模型

    ********* 此处使用的基础镜像为 nvcr.io/nvidia/digits:18.06,镜像大小为6.04GB,可从nvidia官方pull此镜像: 容器配置: CUDA:9.0 CUDNN ...

  9. Nginx-HTTP之框架的初始化

    http 框架的初始化与 nginx-rtmp 框架的初始化类似: Nginx-rtmp之配置项的管理 1. ngx_http_module_t ngx_http_module 核心模块定义了新的模块 ...

  10. storm java.io.NotSerializableException

    今天编写一个storm的topology,bolt的逻辑跟之前的类似. 为了减少重复代码,我建了个抽象基类,存放bolt的公共逻辑,设计了几个abstract方法,不同的逻辑部分由子类实现. 基类日志 ...