1/8
Looks like no tags are added yet.
Name | Mastery | Learn | Test | Matching | Spaced |
---|
No study sessions yet.
Hypothesis class
Can be viewed as reflecting some prior knowledge that the learner has about the task
No Free Lunch Theorem
No learner can succeed on all learnable tasks Every learner has tasks on which it fails while other learners succeed
Error decomposition
Approximation error
Estimation error
Approximation error
Measures how much inductive bias
Estimation error
Derives from inability to choose (with ERM) the best hypothesis
Bias-complexity tradeoff
larger (more complex) H: decreases app err but increases est err → overfitting
smaller H: increases app err but decreases est err → underfitting
Test set
New set of samples not used for picking h_S (=the training set), so to be able to estimate true error