What is candidate sampling?

Candidate sampling is a method used to reduce the computational cost of training machine learning models by selecting a smaller set of examples from a large dataset. This is particularly useful when the dataset is too large to fit into memory or when the model is too complex to train on the entire dataset.

For example, in a recommender system, candidate sampling can be used to select a subset of movies from a large catalog to recommend to a user. The system can select a few movies that are most similar to the user's past preferences, rather than trying to consider every movie in the catalog.

Candidate sampling is used in various machine learning algorithms, such as neural networks and gradient boosting machines, to reduce the computational cost of training. It is particularly useful in natural language processing and recommender systems.

One common misconception is that candidate sampling is a form of data pruning, where the goal is to remove irrelevant examples from the dataset. However, the goal of candidate sampling is to select a representative subset of examples, not to remove irrelevant data.

Candidate sampling has been used in machine learning for several decades, but it has become increasingly important in recent years with the growth of large datasets and complex models.