Online Learning for Changing Environments

Organisations have been gathering large amounts of data. Based on such data, predictive models can be created using machine learning algorithms in order to provide insights into several tasks. For example, data on credit card customers can be used to create credit card approval models for predicting whether a new customer is likely to pay their bills or not. Data on news articles and information on which articles have been read by a customer can be used to create information filtering models for predicting whether an individual would be interested in reading an article. Data on the price of electricity can be used to create models for estimating future electricity price.

One of the key challenges in creating predictive models is that most real world problems change over time. For example, economic crises may change credit card customers' behaviours, news readers may change their reading interests and climate changes may affect the price of electricity. Such changes can make old predictive models inadequate. Therefore, machine learning algorithms need to be able to adapt predictive models to such changes. We have been proposing new algorithms to minimise the negative effects that changes may have over predictive models and speed up the recovery of predictive models from such changes.

Share this page: