After doing a bit of research, I'm somewhat convinced that Hamming networks are networks that classify, but only put out a single result, as opposed to other networks which output multiple results. In this case, is the equivalent in terms of the NEF a classifier (measuring similarity with a dot product between two vectors) followed by an associative memory?

  • $\begingroup$ Nice question, I'll have a look when I'm back home on Sunday. $\endgroup$ – Christian Hummeluhr Apr 8 '15 at 8:05

An Associative Memory is a classifier and is equivalent to a Hamming Network.

For documentation on the NEF Associative Memory, see this practical Nengo documentation and the paper "A biologically realistic cleanup memory: Autoassociation in spiking neurons". Basically, each ensemble of an associative memory computes the similarity measure of the input firing pattern to a given firing pattern. Mutual inhibition is then used between the nodes to find the most similar input so that only the ensemble with the most similar input outputs a signal. In an associative, this output similarity signal is used to map the input pattern to a new output pattern.

A Hamming Network is a classifier that receives patterns with input nodes also detecting similarity. However, unlike the Associative Memory it has a binary output using mutual inhibition. That being said, it would be easy to add a thresholding ensemble to the output of the Associative Memory to accomplish this.

In conclusion, an Associative Memory is basically a Hamming Network implemented using the NEF.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.