Midterm10-701 Fall 2006Outline•Statistics for homework 2•Overview of the midterm•What you need to know•Go over problems from previous midtermsGeneral 58 homework submission including 26 late ones (last time 64 homeworks) Mean: 75.3 (7.9+15.5+30.6+21.2) Stdev: 14.0 Median: 76.5HistogramOverview•Open notes, open book•No electronic devices (computer, cell phone, calculator)•A lot of short problems (varying difficulty, but easier than the homeworks)•Understand materials, look for interesting structures in problemsWhat you need to know•General•Decision trees•Probability, MLE, MAP•Linear regression•Generative and discriminative classifiers (naive Bayes and logistic regression)•Neural network•Model selection•Boosting•SVM, kernel methods•PAC learningGeneral•Training error•Test error•Decision boundaryDecision trees•Concept•ID3•Entropy, conditional entropy•Information gainProbability, MLE, MAP•Axioms of probability•Conditional probability, Bayes rule•Independence, conditional independence•Likelihood, MLE•Prior, posterior, MAPLinear regression•Regression (vs classification)•Estimation (gradient descent, normal equation)•Probabilistic interpretationGenerative and discriminative classifiers•Bayes classifier, Naive Bayes, assumptions•Logistic regression, assumptions, regularization•Relationship•Generative vs discriminative classifiersNeural networks•What is a neural network•What is an activation function•Hidden layer•Backpropagation algorithm•RegularizationModel selection•Purpose•Methods: cross-validation, score•Bias-variance decomposition•Feature selectionBoosting•Ensembles of classifiers•AdaBoostSVM•Margin•Support vectors•KernelPAC learning•Size of hypothesis space•Epsilon, delta•VC-dimension,
View Full Document