Tutor profile: Vardan P.

Explain the difference between test, train and validation sets in supervised learning.

In supervised learning we train our model on the train set and check the performance on the validation set. The goal is to check if the model has overfitted and how it will perform on unseen data. If the results on validation set are much worse, then we have an overfitting problem. However, over time, by performing hyper-parameter tuning, the validation set stops being as "unseen" as we would like it to be, and there arises an opportunity that we overfitted the model for validation set by tweaking the hyper-parameters. So, we have another, third, train set, which is only used at the very end to check the metrics on a truly unseen data.

What is a Turing test?

A Turing test is a way to test the human level intelligence of the machine. During the test, the human interviewer has to identify whether he is confronted with a machine agent or a human agent. The machine will pass the test if the human mistakes it for a human agent.

Briefly explain the curse of dimensionality and suggest a way to fix it.

When we have high dimensional data (i.e many features), many problems arise that are not present in low dimensional space. For example in high dimensional space the distances between two points is very small, and hence most of ML algorithms will not work here anymore. Too fix the issue we can use dimensionality reduction techniques (e.g PCA)