Prüfung Pattern Recognition, SS 2015 (Oktober)

Elmar Noeth

1. Bayesian Classifier
- What is Bayesian Classifier and what is the bayesian decision rule ?
- Can you write down the Bayesian formula ?
- What is p(x/y) ?
- Why do we ignore the evidence p(x)?
- Then he drew two gausian distributions with unequal co-variance matrix and asked me , what would be the shape of the decission boundry ?
- What happens to the decision boundary if the Prior P(y) of one class is higher than the other ?
- What other distributions do we have other than Gausian ?

2. Naive Bayes
- How many parameters do we need to estimate for a 50 dimentional feature vector ?
- What should i do if I want to reduce the parameters to be estimated ?
- What does Naive bayes do ?

3. Support Vector Machines
- What do i do when i have linearly separable cases and I need a unique decision boundary ?
- How does the support vector machine work ?
- Can you write the optimization rule in both hard and soft margin case with constraints ?

- What will happen If i use perceptron in the same very case ?
- When does the perceptron converge ?

4. Gausian Mixture models
- What are gausian mixture models ?
- How do we solve the problem of gausian mixture models ?
- Explain how E-M Algorithm works ?
- What is the constraint of the P(k/y) ? Sum of P(k/y) = 1 ?

5. Adaboost - What is a definition of a weak classifier ?
- What happens in Adaboost ?
- How are misclasified features treated in the next classifier ?
- How is the decission made in the end ?
- There is a modification to Adaboost algorithm , that is used for face detection , what is it called and what does it do ?
- Why does subsequent classifier perform better ? (Because it has more features.)

Overall the atmosphere was very friendly , the professor wants to keeps the discussion very interactive , He likes when you keep on discussing more about the subject and the question asked. Mostly evaluations are done when you use certain key words like a name of a classifier or something .

Note: 1.3