Machine Learning

Learning Model

A mechanism that encodes “knowledge” for problem solving

Cost/Loss (Utility) Function

performance criterion for a specific learning goal based on a given parametric learning model and training data to judge how well the parameters of this model are set based on a training dataset.

Learning Algorithm

optimisation process that minimises (maximises) the cost/loss (utility) function with respect to model parameters

Types

• Supervised ML
• Unsupervised ML
• Semi supervised ML
• Weak Supervision
• Distant Supervision

Assumptions

• IID

Challenges

• training data
  • insufficient
  • nonrepresentative
  • poor quality
  • irrelevant features
• bias-variance tradeoff

Tasks

• Classification
• Regression
• Clustering