Paul Irofti
University of Bucharest & Institute for Logic and Data Science, Bucharest, Romania

Dictionary learning (DL) is a factorization method with many applications to audio and image processing, compression, classification, and computer vision, where it gives better performance than popular transforms. Finding an optimal dictionary that generates at the same time the sparsest representations of data and the smallest approximation error is a hard problem. Our study investigates how the standard DL optimization problem can be modified to perform anomaly detection. We start from a result presented at ICASSP'22 that focuses on uniform sparse representations models that recover the subspace of the majority of samples in a dataset using a K-SVD-type algorithm. Afterwards we continue with on-going work and results in this direction.