Adaptive Resonance Theorem Networks can be seen as an upgrade to self organising maps as the implemented technique addresses the shortcomings of the latter. ART is also a self organising map, however it can switch between learning and fixed states of operation, in order to be able to adapt without losing the memory that has already been stored from past experience. In a way, it can be compared to how brains work. As humans, we are able to learn without losing what we have learned previously. We repeat this process in the course of our lives when we come across new challenges on a daily basis.
ART networks implement a clustering algorithm. Input is presented to the network and the algorithm checks whether it fits into one of the already stored clusters. If it fits then the input is added to the cluster that matches the most else a new cluster is formed. Unsupervised techniques include ART1 (binary input value classification) and ART2 (continuous-valued input data). ARTMAP is an example of supervised machine learning techniques for adaptive resonance theorem.
A research suggests introducing a dynamic clustering algorithm in order to not only prevent discarding irregular data or giving rise to dead neurons but also cluster unlabelled data when the number of clustering is unknown. In the experiments, the same data are used to train the adaptive resonance theory network and the dynamic clustering algorithm network. The results prove that dynamic clustering algorithm can cluster unlabelled data correctly. More information on this research can be found at https://link.springer.com/chapter/10.1007/978-1-4020-3953-9_35