Sur quelques idées fausses ayant des conséquences en identification

Eric Walter (1) , Michel Kieffer (1)

1 L2S - Laboratoire des signaux et systèmes

Identification is more an art form than an exact science, and it is based on intuition just as much as on a large variety of techniques. Tradition is a major factor, and customs that may only have history as a justification should be questioned. Examples are the confidence bestowed on classical mathematical formulas even when they turn out not to be suited for computer implementation, the quasi-exclusive role played by the minimization of quadratic cost functions, the use of finite-difference approximations for the computation of gradients, the almost exclusive resort to local, non-guaranteed techniques for the identification and simulation of nonlinear models. For all of these topics, alternative approaches are suggested.

Domaines: Traitement du signal et de l'image [eess.SP]; Analyse de données, Statistiques et Probabilités [physics.data-an]; Traitement du signal et de l'image [eess.SP]

Eric Walter, Michel Kieffer. Sur quelques idées fausses ayant des conséquences en identification. Journal Européen des Systèmes Automatisés (JESA), 2008, 42 (2), pp.181-210. ⟨hal-00549148⟩