CSC480/680: Midterm Exam
Overview of concepts, algorithms and techniques that you are responsible for (in no particular order) [Please let me know if I have missed anything important!]
1. Concepts that you need to be able to describe and explain: 1.1 General Concepts:
– Optimal Bayes Learning
– Classification, Regression, Concept Learning, Multi-class learning
– Concept, Instances, Classes, Generalization
– Inductive bias
o Representational bias
o Preference bias
– Occam¡¯s Razor
– Version Space
– Overfitting
– Pruning
– Bias, Variance, Bias-Variance Dilemma
– Sample Complexity
– PAC Learning
– Curse of Dimensionality
– Outlier
– Parametric, Non-parametric
– Class Imbalance, Class Skew
– Nominal, Continuous Features, Feature Space, Instance Space
– No Free Lunch Theorem
– Class Noise, Attribute Noise
– Linearly separable, Non-linearly separable
– Linear classifier, nonlinear classifier
1.2 Methods
– Unsupervised Learning, Clustering
– Reinforcement Learning
– Feature Selection: filters, wrappers
– Class Imbalances: undersampling, oversampling, SMOTE
– Deep Learning
2. Algorithms that you need to understand in detail
– Decision Trees
– Multiple Layer Perceptrons
– Bayesian Learning
– Instance-Based Learning
3. Evaluation metrics/methods that you need to understand in detail – Evaluation Metrics:
o true positive, true negative, false positive, false negative, true positive rate, false negative rate
o Accuracy, Error rate
o Precision, Recall, F-Measure
o ROC Analysis, Area Under the Curve
– Re-sampling:
o Cross-validation, Stratified Cross validation, k-fold cross validation o Leave-one-out (Jacknife)
o Bootstrapping
– Statistical Testing:
o General concepts: hypothesis testing, significance test, confidence interval, p-value,
critical value, omnibus test, post-hoc test o T-test
o Sign test
o McNemar test
o Wilcoxon Signed Rank test o ANOVA
o Friedman Test
o Tukey¡¯s test
o Nemenyi¡¯s test
4. From the research paper readings
o Re-read all the abstracts, introductions and conclusions of the papers you have reviewed. You don¡¯t need to know exactly how they do what they do, but you should have a clear enough sense of what they try to do, why it is important, and whether they achieved what they set out to achieve. You can concentrate on the two papers you liked most and the two papers you liked the least and be ready to explain why you liked or didn¡¯t like them. This can be done in comparison with the other papers, but it doesn¡¯t have to be. (Also, note: I am not interested in how they were written: this question has to do with the problems they try to solve, the method they used, etc…)