Model Breaking Detection Using Independent Component Classifier

This paper presents a neural architecture for model breaking detection in real world conditions. This technique use an Independent Component Classiier 1] for detection of unexpected or unknown events in noisy and varying environment. This method is based on subspace classier 2] and Independant Component Analysis 3]. A feed-forward neural network adapts itself to input evolutions, by detecting novelties, creating and deleting classes. A second process achieves a prototype rotation in order to minimise mutual information of dierent classes. This synaptic weight evolution rule is based on an anti-hebbian learning rule inspired from neural methods for blind separation of sources 4]. Consequently , under the assumption of statistical independence of dierent classes, the system is able to detect novelties hidden by simultaneous acoustic events. Novelty detection performances in various situations have been tested : isolated novelty, novelty which occurs mixed with an event of a known class, and several simultaneous novelties. We have also studied the evolution of detection performances obtained by varying the noise level. These experiments have shown good detection performances and low false detection rate.