Types:
- Supervised: labeled training set
- Unsupervised: discover patterns in unlabeled data
- Reinforcement learning: learn to act based on feedback/reward
Deep learning:
- Learning representations of data - great at learning patterns
- Uses a hierarchy of multiple layers - hence the ‘deep’ in the name
- Convolutional neural networks
- Works both supervised and unsupervised
- Compared to machine learning, a lower rate of diminishing returns as the size of the training set increases
Neural networks:
- Input layer
- Hidden layer(s)
- Output layer
Activation functions: non-linearities needed to learn complex (non-linear) representations of data. More layers and neurons can approximate more complex functions.
Overfitting: when the model fails to generalize outside the training set