CONCAVE LOSSES FOR ROBUST DICTIONARY LEARNING

Autor(es)

de Araujo, Rafael Will M. ; Hirata, R., Jr. ; Rakotomamonjy, Alain ; IEEE

• Tipo de documento: Artigo Científico
• Fonte: 2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP); v. N/A, p. 5-pg., 2018-01-01.

Resumo

Traditional dictionary learning methods are based on quadratic convex loss function and thus are sensitive to outliers. In this paper, we propose a generic framework for robust dictionary learning based on concave losses. We provide results on composition of concave functions, notably regarding super-gradient computations, that are key for developing generic dictionary learning algorithms applicable to smooth and nonsmooth losses. In order to improve identification of outliers, we introduce an initialization heuristic based on undercomplete dictionary learning. Experimental results using synthetic and real data demonstrate that our method is able to better detect outliers, and thus capable of generating better dictionaries, outperforming state-of-the-art methods such as K-SVD and LC-KSVD. (AU)

• Processo FAPESP: 15/01587-0 - Armazenagem, modelagem e análise de sistemas dinâmicos para aplicações em e-Science
• Beneficiário: João Eduardo Ferreira
• Modalidade de apoio: Auxílio à Pesquisa - Programa eScience e Data Science - Temático