Title:Unsupervised Unlearning of Concept Drift with Autoencoders

Abstract:The phenomena of concept drift refers to a change of the data distribution affecting the data stream of future samples -- such non-stationary environments are often encountered in the real world. Consequently, learning models operating on the data stream might become obsolete, and need costly and difficult adjustments such as retraining or adaptation. Existing methods to address concept drift are, typically, categorised as active or passive. The former continually adapt a model using incremental learning, while the latter perform a complete model retraining when a drift detection mechanism triggers an alarm. We depart from the traditional avenues and propose for the first time an alternative approach which "unlearns" the effects of the concept drift. Specifically, we propose an autoencoder-based method for "unlearning" the concept drift in an unsupervised manner, without having to retrain or adapt any of the learning models operating on the data.