aka Feature Learning
Data Representation
effective way to encode information underlying data
in humans the information processing is hierarchical.
Acquisition Methodology
- Hand-Crafted feature engineering
- Representation learned from data
Motivation
Curse of Dimensionality Latent Factor in Manifolds
Supervised Representation Learning
mapping from input to targets use deep Neural Networks
Unsupervised Representation Learning
- Dimension Reduction
- Manifold learning
- density estimation/recovering distribution using Generative Models
- Disentangling underlying factors of variation via hybrid learning