c2b_default

KDDM

KDDM@LMU

KDDM@LMU

203
0.0 (0)

Fichier Détails

Cartes-fiches 203
Langue Deutsch
Catégorie Informatique
Niveau Université
Crée / Actualisé 01.08.2019 / 03.08.2019
Lien de web
https://card2brain.ch/box/20190801_kddm
Intégrer
<iframe src="https://card2brain.ch/box/20190801_kddm/embed" width="780" height="150" scrolling="no" frameborder="0"></iframe>
Étudier

Name two famous kernels.

  • polynomial kernel
    • includes correlations = crossterms
  • radial base function (RBF) / gaussian kernel 
    • distance of features non-linear weighted

What are sources for errors? (2) How to tackle them? (2, 3)

  • overfitting = large difference between test-error (large) and train-error (small)
    • reduce VC(H) 
    • increase |S| = m
  • underfitting = large training-error
    • increase VC(H) 
    • increase d (number of features)
    • or: problem not learnable

What are learning curves good for? How do act in both cases?

  • see if train- and test-error are likely to converge
    • if not: reduce complexity of learner
    • if true: increase m for convergence

Étudier