MLSEC 2 Flashcards
Generative learning models
Modeling of underlying concepts
Discriminative learning models
Modeling of prediction function only
Learning process
Finding a “good” learning model in Θ for prediction
Supervised learning
with labels
g : X × Y → Θ
Unsupervised learning
without labels
g : X → Θ
Training error
incorrect predictions on the training data
Test error
incorrect predictions on unknown test data
Loss function
L : Y × Y → ℝ
Comparison of true label with predicted label
Empirical risk Rn
Average loss on training data
Expected risk R_inf
expected loss on data from same distribution
Optimal learning model
min of Expected risk R_inf
What is Classification (Supervised)
Discrimination of objects using learning model (discrete output)
What is Regression (Supervised)
Approximation of observed function through learning model (continuous output)
What is Clustering (Unsupervised)
Contrast to classification: clusters not known at start
What is Anomaly Detection (Unsupervised)
Detection of deviations from learned model of normality
