Towards statistically valid population decoding models

We focus in this paper on the methodology of building statistically valid population code read-out models for spike train data. A new method is explored, which uses Bayesian networks to formalize the read-out model, Monte Carlo validation to check the statistical validity of the model and scrambled quasi-random vectors to speed up the validation process. This procedure avoids imposing usual additional constraints on the data. We present the method through an application in the context of non-metric categorical vision-related data. © 2002 Elsevier Science B.V. All rights reserved.