Bayesian framework applied to a source separation problem using a priori knowledge concerning ill conditioned mixtures issued from an Eddy current sensor response
Michel Haritopoulos · AIP conference proceedings · 2001
In this paper, we deal with the Bayesian framework applied to the source separation problem. Previous work showed that, the incorporation of additional information available from prior experiments or knowledge about the physics of the specific problem, is possible. Motivated by the ill conditioned nature of experimental signals issued from an eddy current sensor nonlinear response, we illustrate how one can incorporate prior information about the mixing matrix using Bayesian formalism. From this, we derive a modified Bell-Sejnowski learning rule. Results of its application to simulated and real world signals are provided.