Types Of Learning (ML)

Supervised vs Unsupervised Learning

Supervised Learning

Supervised Learning is where the agent is given matching pairs of data and conclusions, and is trying to learn a rule to fit the two. It uses concept learning and can use decision trees in ML or artificial neural networks.

Problems

  • Framework (decision tree, artificial neural networks, SVM, etc)
  • Representation (of inputs and outputs)
  • Pre-processing / post-processing
  • Training method (perceptron learning, backpropagation, etc.)
  • Generalisation (avoid over-ļ¬tting)
  • Evaluation (separate training and testing sets)

Unsupervised Learning

Unsupervised Learning is where the agent is given data without any hints and told to find patterns (find attributes which can be grouped together in characteristic "patterns"). It uses clustering and association rule learning in ML to do so.

Reinforcement Learning

Reinforcement Learning is where the agent is not presented with target outputs, but is given a reward signal, which it aims to maximize.

Read more about reinforcement learning

Comparison

  • Supervised Learning is Direct Feedback (through lots of examples)
  • Reinforced Learning is Indirect Feedback (after many examples)
  • Unsupervised Learning is No Feedback

Direct Experience vs Indirect Experience

  • Direct Experience is where each play-out has been done by the agent itself, rather than the knowledge being received secondhand (i.e. from a database). Hence playing out a series of moves and realising they equate to a loss is direct.
  • Indirect Experience is the opposite - e.g. noting that a move leads to a loss based on data provided.

Active Learning vs Passive Learning

(I.e. is it the teacher or not?)

  • Active learning is where the Agent is the Teacher - it is choosing the test data that it feels relevant, rather than following provided data.
  • Passive learning is the opposite, it is being given test data and working through them to build up its knowledge database, but not working out which data will be most useful itself.