10机器学习_课程教案

《10机器学习_课程教案》由会员分享，可在线阅读，更多相关《10机器学习_课程教案（9页珍藏版）》请在装配图网上搜索。

1、Machine Learning Course PlanLecture OneTitle: Introduction Content: l Basic information about this course: books, TA, office, homework, project and test forml Introduce the definition of learning systems. l Give an overview of applications to show the goals of machine learning. l Introduce the aspec

2、ts of developing a learning system: training data, concept representation, function approximation. Targets:l Understand the background of machine learning;l Remember the basic function of machine learning;l Get the general ideas of machine learnings problem and point;Processions:l What is Machine Le

3、arning?l Applications of MLl Disciplines relevant to MLl Well-Posed Learning Problemsl Designing a Learning Systeml Perspectives and Issues In Machine Learningl How To Read This BookDifficulties:l How to design a learning system;l The understand of concept.Lecture TwoTitle: Concept Learning and the

4、General-to-Specific Ordering Contents:l What is the concept learning task? Where is it applied to?l Make students understand concept learning is equivalent to search through a hypothesis space. l Illustrate step by step the procedure of general-to-specific ordering of hypotheses, to find the maximal

5、ly specific hypotheses Targets:l Understand the background of concept learning;l Remember the basic concept of concept learning, version space, etc;Processions:l Introductionl A Concept Learning Taskl Concept Learning as Searchl Find-S：Finding a Maximally Specific HypothesisDifficulties:l Whats the

6、process of concept learning;l Remember the concept of concept learning;Lecture ThreeTitle: Candidate elimination and Inductive bias Contents:l Introduce the definition of version spaces and the candidate elimination algorithm.l How to learning conjunctive concepts?l Introduce and emphasize the impor

7、tance of inductive bias. Targets:l Remember the process of candidate elimination algorithm;l Get the basic idea of the useless of unbiased learning;Processions:l Version Spaces and the Candidate-Elimination Algorithml Remarks On VS and C-El Inductive BiasDifficulties:l The understand of version spac

8、e;l The idea of bias;l The under stand of Find-S Algorithm and Candidate-Elimination Algorithm;Assignments:l EX. 2.1l EX. 2.4Lectrue FourTitle: Decision Tree Learning(1)Contents:l Development of Decision tree learning, the role it plays in the history of increcemental learningl Show the students how

9、 to representing concepts as decision trees.l Introduce recursive induction of decision trees. Targets:l Understand the background of Decision Tree;l Remember the basic concept of decision tree, over fitting, etc;Processions:l Introductionl Decision Tree Representationl Appropriate Problems for Deci

10、sion Tree LearningDifficulties:l One of the most widely used and practical methods for inductive inference l A method for approximating discrete-valued functionsl Robust to noisy datal Capable of learning disjunctive expressionsLectrue FiveTitle: Decision Tree Learning(2)Contents:l Introduce recursi

11、ve induction of decision trees. l Picking the best splitting attribute: entropy and information gain. Emphasize this part, let students do exercise to practice the procedureTargets:l Remember the process of the learning algorithm of decision tree;Processions:l The Basic Decision Tree Learning Algori

12、thml Hypothesis Space Search ID3Difficulties:l ID3, Assistant, C4.5Lectrue SixTitle: Decision Tree Learning(3) Contents:l What is Overfitting? When will is happen? What damage will it cause to classifiers? What should be done in case of noisy data? Why and how to prune?l How to apply the decision tr

13、ee to continuous attributes and missing values. Targets:l Get the basic idea of solving the problems;Processions:l Inductive Bias in Decision Tree Learningl Issue In Decision Tree LearningDifficulties:l Inductive bias is a preference for small trees over large treesl Can also be re-presented as sets

14、 of if-then rulesLectrue SevenTitle: Artificial Neural Networks(1) Contents:l What is Neurons? What is the biological motivation of Artificial Neural Networks?l What are linear threshold units and their functions?l Introduce the principle of perceptrons: representational limitation and gradient desc

15、ent training. Targets:l Understand the background of Neutral network;l Remember the basic concept of neutral network, over fitting, etc;Processions:l Introductionl Neural Network Representationsl Appropriate Problems For Neural Network Learningl PerceptronLectrue EightTitle: Artificial Neural Networ

16、ks(2) Contents:l Introduce the component of an Artificial Neural Networks:l Multilayer networks and backpropagation. Hidden layers and constructing intermediate, distributed representations. Targets:l Remember the process of the learning algorithm of neutral network;Processions:l Multilayer Networks

17、 And The Back propagation Algorithml Notation SpecificationLectrue NineTitle: Artificial Neural Networks(3)Contents:l The problem of Overfittingl How to learn network structure, recurrent networks. l Face Recognition exampleTargets:l Get the basic idea of solving over fitting;l Learn how to slove re

18、al problem with ANN;Processions:l An Illustrative Example: Face Recognitionl Alternative Error FunctionsProjects:l Face Recognitionl CheckerLecture TenTitle: Evaluation HypothesisContents:l Motivation for Evaluation Hypothesis l Estimating Hypothesis Accuracy l Basics of Sampling Theoryl A General A

19、pproach for Deriving Confidence Intervalsl Difference in Error of Two Hypothesesl Comparing Learning AlgorithmTargets:l Given the observed accuracy of a hypothesis over a limited sample of data ,estimate accuracy over additional examples;l Know how to Compare performance of different algorithms;l Un

20、derstand the best way to use those limited data to learn a hypothesis and estimate its accuracy;Processions:l Motivationl Estimating hypothesis accuracyl Sample Error and True Errorl Confidence intervals for Discrete-valued hypothesesl Basics og sampling theoryl A general approach for deriving donfi

21、dence intervalsl Difference in error of two hypothesesl Comparing learning algorithmsrLecture ElevenTitle: Bayesian Learning(1) Contents:l Give a brief introduction to the following definitions and theories: Probability theory, Bayes rule, and MAP concept learning. Naive Bayes learning algorithm. Ta

22、rgets:l Understand the background of Bayes;l Remember the basic concept of Naive Bayes learning algorithm, etc;Processions:l Introductionl Bayes Theoreml Bayes Theorem and Concept Learningl Maximum likelihood and least-squared error hypothesesl Maximum Likelihood Hypotheses For Predicting Probabilit

23、yLecture TwelveTitle: Bayesian Learning(2)Contents:l Give a brief introduction to the following definitions and theories: Bayes nets and representing dependencies. Bayes optimal classifers. Minimum description length principal. Targets:l Remember the process of the Bayes optimal classifers;l Get the

24、 basic idea of Bayes nets;Processions:l Minimum Description Length Principlel Bayes Optimal Classifierl Gibbs Algorithml Nave Bayes Classifierl An Example: Learning To Classify Textl Experimental Resultsl Bayesian Belief Networkl EM algorithmLecture ThirteenTitle: Genetic Algorithms Contents:l Overv

25、iew the theory of Genetic algorithm.l Genetic algorithm provide an approach learning that is based loosely on simulated evolution.l How Hypotheses are described by bit strings whose interpretation depends on the application?l The search begins with a population of initial hypotheses. The next genera

26、tion of population is generated by means of operations such as random mutation and crossoverl Introduce the fitness function. l ApplicationTargets:l Know the importance of GA.l Understand the process of GA, and get the idea of how to interpret the hypothesisl Understand the selection, crossover and mutation. Processions:l Introduction the motivation of GAl A prototypical genetic algorithml Representing of Hypothesesl Genetic operators: selection, crossover and mutationl The fitness function and selection