Sie befinden sich hier: Termine » Prüfungsfragen und Altklausuren » Hauptstudiumsprüfungen » Lehrstuhl 5 » pr-2019-02-19   (Übersicht)

Examiner: Prof. Nöth

Atmosphere: Friendly. He helped me out when I was stuck and didn't get where he wanted to go with his question. He seemed to do the whole exam spontaneously and included some short lecturelike sequences. He graded me 1.0 even though I was stuck some time and gave 1-2 incorrect answers.

Bayes-Classifier

  • decision rule
  • pdf modelling (Gaussian etc.)
  • drawing a decision boundary (offset through priors etc.)
  • relationship to KNN (P(bayesian) ⇐ P(nearest_neighbour)⇐ 2*p(bayesian))
  • naive Bayes (impact on decision boundary)
  • same covariance matrices (linear decision boundary)

LDA

  • steps (joint covariance matrix, SVD, calculation of the transform,
  • apply transform to class means)
  • decision rule

GMM

  • equation (summation of the components)
  • EM algorithm (write down p_ik, µ, etc.)
  • initialization of the EM algorithm (random, k-means, apply repeatedly)
  • drawbacks (local algorithm, slow convergence)

Boosting

  • basic procedure
  • error rate
  • weights of the samples
  • drawbacks of exponential loss (overfitting mislabeled data)