Appendix - Scalable Bayesian GPFA with automatic relevance determination and discrete noise models AFurther analyses of preparatory dynamics in the primate reaching task max sim

Here we briefly consider why introducing a prior over the factor matrix enables automatic relevance determination. These ideas reflect results by Bishop [1] and our experiments in Section 3.1. For simplicity, we will first consider the case of factor analysis where p(X) = Q d,tN(xdt; 0,1).