Supervised learning framework

In the previous steps, we introduced a framework for both regression and classification. In this video, let me summarize the general supervised learning framework and provide a comparison between the regression methods and classification methods.

The main objective of supervised learning is to learn the functional relationship between the input and output from the labelled datasets. Depending on whether the output variable is categorial or continuous, supervised learning can be further divided into

• regression;
• classification.

To tackle a general supervised learning task, the main stages include

(1) specifying the input-output pairs (Dataset),

(2) choosing a model to approximate the estimated output (Model);

(3) defining a loss function to quantify the discrepancy between the observed output and the model estimated output (Empirical Loss);

(4) learning optimal model parameters by minimizing the loss function via a suitably chosen optimization method (Optimization);

(5) making the prediction on any given input using the trained model (Prediction);

(6) computing the test metrics on the test dataset in order to assess the model performance (Evaluation).

You may use the above framework to learn any new supervised learning algorithm. If you are able to answer each step of a supervised learning algorithm, you will have a good understanding of the new algorithm.