Incremental Learning

A model trained once goes stale as new input keeps arriving; incremental learning extends it instead. The page gives the definition and two diagrams, then streaming examples in Weka, and ends with the Hoeffding tree.

The same notes are in Follow the regularized leader, Online Learning, and Training Strategies. The same notes are in Decision Trees.

The definition comes from the wiki: in computer science, incremental learning is a method of machine learning in which input data is continuously used to extend the existing model's knowledge i.e. to further train the model. The two diagrams below draw that loop; both were copied from the original hosted image.

Incremental learning.

Incremental learning.

The natural place to use it is a stream of time series. Using incremental learning for time series, i.e., sentiment from twitter api stream, learning from smiley in weka. is the first example from that playlist. And time series in weka is the companion video, with some ideas about dealing with time series data, especially date and time.

Hoeffding tree

The streaming examples still need a model to extend, and the one this page names is the Hoeffding tree. The same notes are in Decision Trees.

The note on it is short: it IS STATE OF THE ART.