This tool is used to infer the posterior probabilities of unknown variables given known values for other variables. This involves normalizing the Bayesian probabilities of the unknown variables so that they sum to 1. We can infer the posterior probability of any of the possible values of the Combination given known values for all other variables by normalizing the Bayesian probabilities for all values of Combination and the known variables. This inference specifies the probabilistic relationships among the variables in a model but for large models with many variables, computation of the Bayesian Inference is infeasible. The dependence of the location and feature also has important implications for probabilistic inference using the model. The enhancement of area V4 neuron responses due to increased attention to a different location, can both be simulated by the image-generation model of bottom-up/top-down interactions in the visual system. This image-generation model can be used to simulate effects in which the response of an area V4 neuron is selective for something over another. The "attentional effect" is interpreted as an increase in the posterior probability brought by an increase in the expectation of the location. Simulations involve larger set of images and combinations than are considered in the models, however the basic modeling approach is the same. The difference between the tools of BUTDjointDistribution and BUTDprobInference are the methods in which the posterior probabilities are computed. For BUTDprobInference, it is through the Bayesian Inference.
Anastasio, Thomas J. Tutorial on Neural Systems Modeling. Sunderland: Sinauer Associates, 2010. Print.
Researchers should cite this work as follows: