Predictive coding is both a technique for efficient information encoding and a method for performing perceptual inference. It is commonly used to model information processing in the cerebral cortex.
Predictive coding models of cortical function are typically implemented as hierarchical neural networks. Such networks contain alternating populations of “error-detecting” neurons and “prediction” neurons. Figure 1 shows the connectivity between a single pair of adjacent populations in such a hierarchy, for a very simplified case where there are only two neurons in each population. The inputs come from the thalamus or are the outputs of prediction neurons at preceding stages in the hierarchy. The activity of the prediction neurons encodes hypotheses about the causes underlying the inputs to the preceding population of error-detecting neurons. The activity of the error-detecting neurons encodes the discrepancy (or residual error) between the expected inputs...
KeywordsResidual Error Independent Component Analysis Perceptual Inference Nonnegative Matrix Factorization Predictive Code
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
- Clark A (2013) Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behav Brain Sci. doi:10.1017/S0140525X12000477Google Scholar
© Springer Science+Business Media New York 2014