Week 7 Flashcards
Feature engineering - modify
Modify measurements to make more useful for classification, eg 09/08/2019 -> Friday
Benefits of feature selection
Reduce risk of overfitting
Reduces training time
Feature extraction
Find features that are functions of raw data
Feature extraction example
KLT
PCA method
NNs for PCA
Oja’s rule
Sanger’s rule
Oja’s subspace rule
How can Auto encoder do PCA
Whitening transform
Because PCA is unsupervised ?
It may eliminate discriminative dimensions
Linear Discriminant Analysis
Independent component analysis
Compare independence and correlation for PCA ICA
NNs for ICA
Random projections
Extreme learning machines
Differences and similarities between random projections and sparse coding
How to - sparse coding
Sparse coding - dictionary based
Generally, dictionary based methods
