论文阅读 | Tackling Adversarial Examples in QA via Answer Sentence Selection
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
- [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 ...
- [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 ...
- 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 ...
- 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 ...
- 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 ...
- 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 ...
- 《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 ...
- 【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 ...
- 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
- 解决从其他地方拷贝过来的Android项目在本机不能运行(报错)的问题
这个问题一般是由gradle版本不同引起的.要解决可以这样: 一.在确保本机Android studio 正常使用的情况下,在本机新建一个Android项目 二.在文件夹中打开新建的Android项目 ...
- javascript内置对象:Date
JavaScript内置函数:Date时间<script> var today=new Date(); weeks=["日","一",& ...
- luogu 1169 [ZJOI2007]棋盘制作 悬线dp
悬线法,虽然得不到局部最优解,但是一定能得到全局最优解的算法,十分神奇~ #include <cstdio> #include <algorithm> #define N 20 ...
- 小米oj 判断是否为连乘数字串
判断是否为连乘数字串 序号:#32难度:非常难时间限制:1000ms内存限制:10M 描述 给出一个字符串S,判断S是否为连乘字符串. 连乘字符串定义为: 字符串拆分成若干数字,后面的数字(从第三个 ...
- HZOJ 20190819 NOIP模拟26题解
考试过程: 照例开题,然后觉得三道题都挺难,比昨天难多了(flag×1),T1 dp?T2 数据结构? T3 dp?事实证明我是sb然后决定先搞T2,但是,woc,这题在说什么啊,我怎么看不懂题啊,连 ...
- 【8.27-模拟赛】remove
题解: 代码: #include<iostream> #include<algorithm> #include<cstdio> #include<cstrin ...
- 编译参数(-D)
程序中可以使用#ifdef来控制输出信息 #include<stdio.h> #define DEBUG int main() { ; ; int sum = a + b; #ifdef ...
- [mysql]my.cnf在哪里
linux mysql --help|grep my.cnf 查找顺序:/etc/my.cnf /etc/mysql/my.cnf /usr/local/mysql/etc/my.cnf ~/.my. ...
- codeforces#1157D. Ehab and the Expected XOR Problem(构造)
题目链接: http://codeforces.com/contest/1174/problem/D 题意: 构造一个序列,满足以下条件 他的所有子段的异或值不等于$x$ $1 \le a_i< ...
- AGC033D Complexity
题意 给出一个\(n*m\)的\(0,1\)矩阵,若一个矩阵中的所有元素都相同,则这个矩阵的代价为\(0\),如果不是则选择一种将它分成两个子矩阵的方案,代价为所有方案中(两个子矩阵的代价的较大值+\ ...