Derivations for the chunking model (Q689455)

TODAM (a theory of distributed associative memory) is a general theory of memory designed to account for the storage and retrieval of items, associative and serial-order information. The chunking model is designed to provide a possible description in a TODAM framework. Some preliminaries necessary for the derivations are provided. The basic concepts of the chunking model are multiple convolutions (convolutions of several item vectors), autoconvolutions of sums of item vectors (\(n\)- grams), and chunks (sums of \(n\)-grams). The values of the model parameters are computed as a function, in order to make predictions or fit data on recognition and recall. Then these expressions are derived first for the variance and then for the expectation. A number of principles resulting from the analysis are listed.