Recurrent autoassociative networks and holistic computations

The paper presents an experimental study of holistic computations over distributed representations (DRs) of sequences developed by the Recurrent Autoassociative Networks (KAN). Three groups of holistic operators are studied: extracting symbols at fixed position, extracting symbols at a variable position and reversing strings. The success with those operators and the very good generalization pave the road for holistic linguistic transformations. Also, it brings a better understanding of the structure of the DRs developed by the RANs.