Algorithms that detect anomalies have to learn normal behaviour to be able to identify anomalous behaviour. If broad features of the expected anomalies are known the use of supervised Machine Learning (ML) is in order. But by definition, the most interesting anomalies are those unexpected, and in that case, unsupervised ML should be used. However, unsupervised strategies are substantially less powerful than possible supervised methods –a catch 22 situation.
A new Machine Learning algorithm called Anomaly Awareness has been developed by Veronica Sanz (IFIC) and Charanjit K. Khosa (University of Sussex). By making the algorithm aware of the presence of a range of different anomalies, the authors improve its capability to detect anomalous events, even those it had not been exposed to. Anomaly Awareness is a new type of semi-supervised learning procedure, applicable to input images but also to other types of information.
As a proof-of-concept, this method has been applied to searches for new phenomena in the Large Hadron Collider (LHC) of the European Centre for Particle Physics (CERN). In particular, events with boosted jets were analysed where new physics could be hiding. The simulation of the LHC environment and the training of the algorithm in various situations have been accelerated by using ARTEMISA.