ML Learning Paradigms

Machine Learning Paradigms

Understanding different learning paradigms is crucial for identifying safety risks.

Supervised Learning

• Learning from labeled examples
• Classification vs regression
• Training, validation, and test sets
• Safety risks: Distribution shift, adversarial examples

Unsupervised Learning

• Finding patterns without labels
• Clustering and dimensionality reduction
• Autoencoders and representation learning
• Safety risks: Hidden biases, unexpected clusters

Reinforcement Learning

• Learning through interaction
• Reward functions and policies
• Exploration vs exploitation
• Safety risks: Reward hacking, unsafe exploration