Multi-layer Sparse Matrix Factorization

Luc Le Magoarou (1) , Rémi Gribonval (1)

1 PANAMA - Parcimonie et Nouveaux Algorithmes pour le Signal et la Modélisation Audio

—The applicability of many signal processing and data analysis techniques is limited by their prohibitive computational complexity. The cost of such techniques is often dominated by the application of large linear operators. This short paper introduces an algorithm aimed at reducing the complexity of applying such operators by approximately fac-torizing the corresponding matrix into few sparse factors. The proposed approach, which relies on recent advances in non-convex optimization, is first exposed, and then demonstrated experimentally.

Domaines: Traitement du signal et de l'image [eess.SP]

Luc Le Magoarou, Rémi Gribonval. Multi-layer Sparse Matrix Factorization. SPARS 2015 Signal Processing with Adaptive Sparse Structured Representations, Jul 2015, Cambridge, United Kingdom. ⟨hal-01158057⟩