Associative memories are a well studied type of neural network model, one of the most famous being Kanerva's Sparse Distributed Memory. I will present a variant of SDM, which is capable of storing *rank-order* information---that is, information that is represented by a permutation over a set of symbols. This is interesting both from a biologically plausible perspective, and also as an interesting mathematical model for information storage.
This is joint with Steve Furber at Manchester, and part of the ongoing SpiNNaker project, into the use of neural-inspired components to engineer more fault tolerant electronics. The speaker is from a Machine Learning background and will not pretend to understand about Computer Architecture - but will do his best.