人工智能10学习1

1、第10章学习大纲 Introduction to machine learning Supervised learning（监督学习） Decision tree learning（决策树学习） Linear predictions （线性预测） Support vector machines （支持向量机） Unsupervised learning （无监督学习）Learning 学习对认识未知的环境是必不可少的, i.e., 当设计者缺乏完整知识的时候when designer lacks omniscience（全知） 学习作为一种系统构造方法是有用的, i.e., 编写一个学习智能体

2、程序来实现比尝试着将函数直接写出来要好得多 学习会不断地修改智能体的决策机制来提高性能学习智能体学习元件Design of a learning element is affected by性能元素的哪些组件是需要学习的什么样的反馈可被用来学习这些组件这些组件可以被哪些方法来表示Introduction to Machine LearningMachine Learning Everyday:Search EngineMachine Learning Everyday:Spam Detection （垃圾邮件检测）Machine Learning Everyday:Machine Transl

3、ationMachine Learning Everyday:Face Detection Now in most digital cameras for auto focusingMachine Learning 脱胎于人工智能工作 电脑的一种新能力Why Machine Learning? Solve classification problems Learn models of data (“data fitting”) Understand and improve efficiency of humanlearning Discover new things or structures

4、 that are unknown to humans (“data mining”)Why Machine Learning? Large amounts of dataWeb data, Medical data, Biological data 昂贵的手工分析费用 计算机变得便宜并性能更加优良Why Machine Learning? 应用程序无法直接通过手工编程完成无人驾驶手写识别自然语言处理(NLP)计算机视觉 理解人类的学习（人脑，真正的AI）What is machine learning useful for? 机器学习在哪些领域有用？Automatic speech reco

5、gnition自动语音识别 当前大部分语音识别和翻译都能够不断学习 你用得越多，它们就会变得越聪明Computer vision: e.g. object, face andhandwriting recognitionInformation retrieval信息检索对大量文本数据库的阅读, 领会和分类对于人类来说是困难的Web PagesRetrieval（检索）Categorization（分类）Clustering（聚类）Relations between pagesFinancial predictionMedical diagnosis（医学诊断）Bioinformatics（生物

6、信息学） e.g. 基因微阵列数据建模, 蛋白质结构预测Robotics机器人学电影推荐系统Movie recommendation systemsMachine LearningMachine learning is an interdisciplinary field focusing on both the mathematical foundations and practical applications of systems that learn, reason and act.机器学习是一个交叉学科的领域，着重于研究具有学习、推理和行动的系统所需要的数学基础以及实际应用Other

7、 related terms: Pattern Recognition（模式识别） , Neural Networks（神经网络） , Data Mining（数据挖掘） , Statistical Modeling（统计模型） .Using ideas from: Statistics, Computer Science, Engineering, AppliedMathematics, Cognitive Science（认知科学） , Psychology（心理学） ,Computational Neuroscience（计算神经学） , EconomicsThe goal of the

8、se lectures: to introduce important concepts（概念）, models and algorithms in machine learning.Machine Learning: 定义 Tom Mitchell (1998) Well-posed Learning Problem: A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T,

9、 as measured by P, improves with experience E. 汤姆米切尔（1998）很好地定义了学习问题：我们说一个计算机程序能从经验E中学会针对某些任务T和一些性能指标P的方法，如果程序使用E有效提高了在T中运行时的指标P. “A computer program is said to learn from experience E withrespect to some task T and some performance measure P, if itsperformance on T, as measured by P, improves with

10、experience E.”假设你的邮件程序观测到你将一些邮件标记为垃圾邮件，以此为基础程序学习如何更好地过滤垃圾邮件，那么在该设定中task T是什么？将邮件分类为垃圾或非垃圾邮件Watching you label emails as spam or not spam.The number (or fraction) of emails correctly classified as spam/not spam.None of the abovethis is not a machine learning problem.学习的种类想象一下，一个智能体或机器收集到一系列的传感输入（sens

11、ory inputs）:x1 , x2, x3, x4, . . .Supervised learning（监督学习） :The machine is also given desired outputs y1 , y2, . . .,and its goal is to learn to produce the correct outputgiven a new input.Unsupervised learning（无监督学习） :outputs y1 , y2, . . . Not given, the agent still wants tobuild a model of x tha

12、t can be used for reasoning,decision making, predicting things, communicating etc.Semi-supervised learning （半监督学习）Representing “objects” in machinelearning 举个实例, x, represents a specific object x 通常表示一个d维的特征向量 x = (x1, . . . , xd) Rd 其中每一个维度叫做feature or attribute 特征值是连续的或离散的 x 在d维的特征空间中是一个点 目标抽象化. 忽

13、略其他方面(e.g., two people having the same weight and height may be considered identical)Feature vector representation特征向量表示法 文本文件词汇 of size d (100,000) “bag of words”: 对每个词条的计数通常忽略掉stopwords: the, of, at, in, 特别的，用 “out-of-vocabulary” (OOV) 来捕捉 所有未知的词 特征向量表示法 图像 像素, 颜色直方图 银行账户 信用等级, 余额, 最近一天、一星期、一个月、一年

14、存款, #取款, You and me 医学特征 test1, test2, test3, 主要成分DataThe data set D consists of N data points:D = x1 , x2 . . . , xNPredictions（预测）We are generally interested in predicting something based on the observed dataset.基于已观测到的数据集能否正确对后来的数据进行预测Given D what can we say about xN+1 ?ModelTo make predictions, we need to make some assumptions. We can often expressthese assumptions in the form of a model, with some parameters（参数）为了完成预测任务，我们需要做一些合理假设。我们经常以带参数的模型形式来表达这些假设在给定数据集D时, 我们学习模型的参数, 以便对新数据进行预测.主要成分学习的架构实验3：Learning ProblemsHousing price predictionSupervised Learning监督学习数据中给出了“right