Supervised ML - given a training set, use a machine learning algorithm to produce a model which can classify things.

• Problem of finding a function for a sample set of inputs/outputs
• 2 phases:
  • Training (offline)
    • Given training data + labels, feed to ML algorithm
    • “Supervised” - needs human labelling (supervision)
  • Testing/deployment
    • Get model from previous step
    • Should be able to take new data and make predictions about it

Note: regression is a form of ML where output is a continuous value.

Classification essentially involves the following problem:

• Input: Object
  • e.g. email
• Output: Label(s) for object (discrete set)
  • e.g. spam? not spam?

Applications of this include:

• Spam Detection
• Sentiment Analysis
• Topic Classification
• Document Ranking
• Object Recognition
• Fraud detection
• NSFW filter

Object Representation (Features)

Computers do not understand lots of data initially. Thus, we need some representation of objects for the algorithms.