Explaining predictive models: the Evidence Counterfactual

Imagine being targeted with an advertisement for this blog. You’d like to know: why did the AI model predict you’d be interested in the Faculty of Business and Economics’ blog, based on the hundreds of web pages you visited? The answer could be: because you visited www.great-business-faculties.com, www.datascience-in-business.org and www.living-in-antwerp.com: if you would not have visited these pages, you’d no longer be targeted with this specific ad. This explanation is an example of an imaginary “Evidence Counterfactual”.

The use of such browsing data, as well as Facebook likes or location data is often used in targeted advertising systems, and beyond. The predictive models are increasingly complex, while end users become more vocal in their wish for transparent and explainable systems.

When being the subject of such predictive models, businesses and citizens might ask: why I am being rejected for credit? Why I am being targeted with this ad? Etc. The European GDPR even provides the “right to obtain meaningful information about the logic involved” to data subjects who are involved in automated decision making.

In this post, you’ll learn more about the Evidence Counterfactual, an increasingly important approach within the “explainable AI” research domain, that helps understand the decisions of predictive systems that use Big Data. The Applied Data Mining research group has developed algorithms to provide such explanations and validated them in a variety of business domains.

Behavioral big data

More and more companies are tapping into a large pool of humanly-generated data, or “behavioral big data”. Think of a person liking Instagram posts, visiting different locations captured by their mobile GPS, searching Google, making online payments, connecting to people on LinkedIn, and so on. All these behavioral traces lead to artificial intelligent (AI) systems with very high predictive performance in a variety of application areas, ranging from finance to risk to marketing.

The goal of these AI systems is to use this data to predict a variable of interest, for example, a person’s personality traits, product interests, creditworthiness, and so on. The model uses a large number of small pieces of evidence to make predictions. Let’s refer to all that data as the “evidence pool“. The pieces of evidence are either “present” or “missing”. All pieces that are present can be used to make predictions.

Tourist or citizen?

To illustrate how behavioral big data can be seen as a “pool of evidence,” imagine a model that uses location data of people in New York City to predict if someone is a tourist or a citizen. Out of all possible places to go to (the “evidence pool”), a person will only visit a relatively small number of places each month.

These are the pieces of evidence that are “present” and are represented by a value of 1 (see Figure 1). All places that are not visited by that person are “missing” and get a corresponding zero value in the data matrix.

In Figure 1, for example, Anna visited 85 places out of the 50,000 possible places used by the model. She visited Times Square and Dumbo, but she did not visit Columbia University, making this a missing piece of evidence. The model decides she’s a tourist.

The intuition behind the Evidence Counterfactual

Explaining how predictive systems make decisions based on big data is challenging. Evidence Counterfactuals helps understand the reasons behind individual model predictions. This explanation approach3 identifies a causal relationship between two events: event A causes Event B, only if we observe a difference in B after changing A while keeping everything else constant.

The Evidence Counterfactual shows a subset of evidence (event A) that causally drives the model’s decision (event B). We imagine two worlds, identical in every way up until the point where the evidence set is present in one world, but not in the other. The first world is the “factual” world, and the unobserved world is the “counterfactual” world. To help clarify this, consider the following:

IF Anna did not visit Times Square and Dumbo, THEN the model’s prediction changes from tourist to NY citizen.

The pieces of evidence {Times Square, Dumbo} are a subset of the evidence of Anna (all the places she visited) and explain the model’s decision. Simply removing Times Square or Dumbo from her visited locations would not change the predicted class. Both locations need to be “removed” (feature value set to zero) to change the model’s decision.

The “factual world” is the one that’s observed and includes all the places Anna visited. The “counterfactual world” that results in a predicted class change is identical to the factual world in every way up until the two locations Times Square and Dumbo.

An important advantage of counterfactuals is that they do not require all features that are used in the model (the “evidence pool”) or all the evidence (e.g., all places Anna visited) to be part of the explanation. This is especially interesting in the context of humanly-generated big data: it allows us to explain predictions using concise and comprehensible explanations.

Computing Evidence Counterfactuals

The huge dimensionality of the behavioral data makes it infeasible to compute counterfactual explanations using a complete search algorithm (this search strategy would check all subsets of evidence as candidate explanations).

Alternatively, a heuristic search algorithm can be used to efficiently find counterfactuals. One existing approach is based on a best-first search and makes use of the model’s scoring function to first consider subsets of evidence that, when removed, reduce the predicted score the most in the direction of the opposite predicted class.

There are at least two weaknesses of this strategy:

  1. for some nonlinear models, removing one feature does not result in a predicted score change, which results in the search algorithm picking a random feature to expand in the first iteration;
  2. the search time is very sensitive to the size of the counterfactual explanation: the more evidence that needs to be removed, the longer it takes the algorithm to find the explanation.

As an alternative to the best-first search, we proposed a search strategy that chooses features according to their overall importance for the predicted score.5 The idea is that the more accurate the importance rankings are, the more likely it is to find a counterfactual explanation starting from removing the top-ranked feature up until a counterfactual explanation is found. The hybrid algorithm LIME-Counterfactual (LIME-C) seems a favorable alternative to the best-first search because of its good overall effectiveness and efficiency.

Other data and models

Evidence Counterfactuals can address various data types, from textual data and tabular data (e.g. standard Excel files) to image data. The issue is to define what it means for evidence to be “present” or “missing.” To compute counterfactuals, we need to define the notion of “removing evidence” or setting evidence to “missing.” In this post, we focused on behavioral big data. For these data, which is very sparse (a lot of zero values in the data matrix), it makes sense to represent evidence that’s “present” by those features (e.g., word or behavior) having a nonzero value.

Key takeaways

  • Predictive systems that are trained from humanly-generated Big Data have high predictive performance, however, explaining them is challenging.
  • Explaining data-driven decisions is important for a variety of reasons (increase trust and acceptance, improve models, gain insights, etc.), and for many stakeholders (data scientists, managers, decision subjects, etc.).
  • The Evidence Counterfactual is an explanation approach that can be applied across many relevant applications and highlights a key subset of evidence that led to a particular model decision.

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处:http://xiahunao.cn/news/353606.html

如若内容造成侵权/违法违规/事实不符,请联系瞎胡闹网进行投诉反馈,一经查实,立即删除!

相关文章

记录搭建hadoop集群的过程

Linux(CentOS-7.6-x64位)基础配置, 虚拟机平台VmWare15 CentOS-7.6-x64镜像下载: https://www.aliyundrive.com/s/72Xg449t6i8 提取码: 32rm VmVare15安装包下载带序列号:VmVare15安装包下载带激活序列号资源-CSDN文库 点击关闭,点击完成&…

ChatGPT热度不减!华为宣布入局,盘古GPT能否大杀四方!

ChatGPT热度不减 六月份了,朋友们,来到六月份了已经,ChatGPT的热度依旧不减,各大论坛网站的榜单上还飘着ChatGPT相关话题的文章,且排名靠前。由此可见,这ChatGPT这股子热潮还得持续一段时间呢。 而且ChatG…

三国志战略版:Daniel_新晋减伤战法-奇计良谋

一、战法介绍 战法定位:稳定且有针对性的群体减伤战法。 同类效果: 稳定的群体减伤战法:八门金锁阵、抚辑军民、暂避其锋、锋矢阵、藤甲兵、御敌屏障。 稳定群体减伤战法往往可以极大提升一个队伍的强度,如今一线群体减伤战法中…

英雄打怪

英雄类 package LOLDemo;/*** ClassName: Hero* Description: * 自定义一个英雄类* 0.英雄类中加入“经验值”,“等级”,“攻击力”等属性变量,加入“击打”方法,击打的目标是怪物,击打的结果是怪物的生命值-英雄攻击力* …

【游戏逆向】RPG游戏自动打怪之取得最近怪物

RPG游戏辅助想要自动打怪,首要的任务就是需要获得最近的怪物,否则打怪的顺序会非常混乱。 想要取得最近怪物信息 要2个步骤 第一步 计算出所有怪物的距离 第二步 对所有怪物的距离进行判断,筛选出距离最近的 第一步怪物距离计算 如果是以2D的平面来看 2个点的距…

英雄杀-如何通关挑战

对于很多英雄杀的新手玩家来说,挑战,几乎很难一次通关,我也是,一般没超过十关。 我给大家分享一下我最近的通关方法,仅供参考: 第一种思路: 杨延昭: 有两个前提: 第一、…

javaweb session与cookie

Cookie Cookie是本地的存储数据,是键值对 注意一定要使用resp.addCookie才能返回cookie cookie修改 修改的话也是调用addCookie,对相同key的进行修改,作用于set-cookie响应头。 可以新建cookie,add进去,也可以在用户传…

行为型设计模式03-观察者模式

🧑‍💻作者:猫十二懿 🏡账号:CSDN 、个人博客 、Github 🎊公众号:猫十二懿 观察者模式 1、观察者模式介绍 观察者模式是一种行为型设计模式,也被称为发布-订阅模式,它定…

【虹科案例】虹科数字化仪在激光雷达大气研究中的应用

01 莱布尼茨研究所使用激光雷达进行大气研究 图 1:在 Khlungsborn 的 IAP 办公室测试各种激光器 大气研究使用脉冲激光束通过测量大气中 100 公里高度的多普勒频移和反向散射光来测量沿光束的温度和风速。返回的光信号非常微弱,会被阳光阻挡&#xff0c…

XML入库后空白字符丢失问题

最近项目上在做电子病历,使用的是第三方的电子病历组件,该病历组件是利用XML来组织数据的。界面上渲染出来的效果如下图: XML渲染后的界面 对应的后台数据(已做简化处理)是如下XML格式的,其中的空格部分是…

前端入门学习

封装axios axios的基础使用 axios基础使用方法: axios.create({config}) //创建axios实例 axios.get(url,{config}) //get请求 axios.post(url, data,{config}) //post请求 axios.interceptors.request.use() // 请求拦截器 axios.interce…

赛尔号什么时候支持html5,赛尔号:这几只精灵必须拥有!无关强度,只因经典情怀!...

赛尔号这款游戏有着比较老的资历,很多玩家从还是小学生的时候就开始玩起了这个游戏,然后便一直玩到大学,甚至毕业工作。因此,赛尔号不仅仅是一款简单的网页游戏,更是一款充满了情怀的游戏。 那么,接下来的小…

赛尔号星球大战服务器维修,赛尔号星球大战11月29日更新公告

赛尔号星球大战11月29日新增了什么玩法?服务器的维护时间是多少?来看看9k9k小编rayxx带来的赛尔号星球大战11月29日更新公告。 11月29号下午15点至17点更新公告: 1、精灵 新增了两种精灵:亚兰洛(火系),星吉拉(普通系)。 通过有缘人进行精灵繁…

4399知名游戏-赛尔号图鉴的爬取

4399知名游戏-赛尔号图鉴的爬取 面向对象: 1.疫情居家无聊之人 2.python略懂一点点就行 头文件引入: 如果没有下面的头文件不要慌,打开你的python终端pip install 包名即可 from bs4 import BeautifulSoup import requests import json …

容器技术 — Cgroups 与 Namespaces 支撑实现的操作系统虚拟化

目录 文章目录 目录操作系统虚拟化(容器技术)的发展历程ChrootCgroupsCgroup SubsystemsCgroup FilesystemCgroup HierarchyCgroups 的操作规则Cgroups 的代码实现 NamespacesUTS namespacePID namespaceIPC namespaceMount namespaceNetwork namespaceU…

使用geoserver发布shp和tiff数据

一、安装并启动geoserver服务 1.1 下载geoserver 进入官网下载 由于geoserver是使用Java语言开发的,所以运行需要java的环境,不同geoserver的版本号对java的版本要求不同,所以选择版本时需注意对应java的版本要求,由于我本地安…

javascript回到顶部

如图: 数字从1到100,滚动后点击章鱼哥便可以回到顶部。 HTML内容不多只有一个a标签内容有一个图片和100个h1标签,但要看清楚是给它设置好了id的,当然不设置id也行但有时候对小白不太友好。 为了使点击图片更加贴合整个页面所以需要…

一段简单的网页返回顶部和返回底部代码(html+css+jquery)

最近用到了返回顶部和底部功能&#xff0c;找了找&#xff0c;查了查&#xff0c;改了改&#xff0c;最终实现效果大概这样子。 以下是相关代码&#xff1a; html部分&#xff1a; <div id"backtotop" class"backtotop"><div class"bt-box…

网页回到顶部的js代码实现

背景 还是给老板写他的个人主页&#xff0c;用到了一个小功能&#xff0c;随手记录下 代码 控制样式的 css 代码如下 /* 回到顶部实现代码的css */ .back-to-top {display: none;/* 默认是隐藏的&#xff0c;这样在第一屏才不显示 */position: fixed;/* 位置是固定的 */bott…

HTML5 回到顶部

图片&#xff1a; html <!DOCTYPE html> <html><head><meta charset"utf-8" /><title>回到顶部</title><link rel"stylesheet" type"text/css" href"css/try.css" /><script src"j…