An article from NewScientist that was published several years ago (May 2004) suggests that a small-world type of network arrangement may be present in the connections between neurons in the areas of the prefrontal cortex that are responsible for short-term memory. The article explains that recent research has suggested that these areas have bistable states of activation–they can be activated into two (or possibly more) distinct self-sustaining patterns of activation. Computational neuroscientists have been struggling to model how this sort of neural activity might occur, and many models of the phenomenon use extremely complex models involving complicated neurochemistry and lots of different types of neurons.

More recently, however, simpler models have emerged suggesting that short term memory could be achieved without so many complicated parameters. These newer models have architectures similar to some of the small-world networks we’ve discussed in class: each neuron connects not only to the neurons around it, but also connect to a few far-away neurons selected at random. These models have been able to exhibit the requisite bistable behavior for short-term memory without much trouble, and similar models lacking the random connections (non-small-world networks) cannot be made to exhibit the same behavior.

One study mentioned in the article found that when 10-20 percent of the neurons used their “short-cut” connections to random distant neurons, self-sustaining bursts of activity would occur. Furthermore, when the activity propagates back to its source, it has an inhibitory effect, and causes the pattern of activation to cease after a short period. This is the expected behavior of a bistable network: it stays in the necessary state just long enough to adequately represent a memory, and then can be quickly cleared away in order for a new to occur in the same place.

Whether such models are biologically plausible is still open for debate, but it is nonetheless interesting to consider applications of small-world networks in this sort of context.