Machine Learning with Swift

Cold start problem

A very common situation is when a machine learning system starts functioning in a new environment, where no information to pre-train is available. The situation is known as a cold start. Such a system requires a certain amount of time to collect enough training data, and start producing meaningful predictions. The problem often arises in the context of personalization and recommender systems.

One solution for this it is so-called active learning, where the system can actively seek new data that could improve its performance. Usually, this means that the system queries a user to label some data. For instance, the user can be asked to provide some labeled examples before the start of the system, or the system can ping him ...