User:H. Sebastian Seung/Proposed/Persistent activity
This is just a placeholder.
Neural activity related to short-term memory has been recorded in a wide variety of brain areas, chiefly in awake behaving primates. (citations needed here) Typically this activity is evoked by a transient sensory stimulus, and persists as an elevated or suppressed discharge for up to many seconds until a behavioral response (figure needed here). As many studies of neural coding have shown, this persistent neural activity is correlated with information stored in short-term memory during the time interval separating stimulus from response.
Both cellular and circuit mechanisms for persistent neural activity have been hypothesized. The most popular candidate for a cellular mechanism is bistability due to plateau potentials or synaptic properties. Since plateau potentials last for seconds, and can be turned on or off rapidly by a transient input, they seem ideal for supporting persistent neural activity. The most popular candidate for a circuit mechanism has been reverberating activity in synaptic feedback loops, as hypothesized by Hebb and Lorente de No. This hypothesis has been implemented in a class of models known as attractor neural networks, which have multiple stable states corresponding to persistent activity patterns.
Both of these hypotheses (multistable cells and multistable circuits) are well-known and appealing in their simplicity. Some neuroscientists casually accept one or the other as true. It is less widely known that both hypotheses, at least in their simplest formulations, are problematic. Both have trouble accounting for graded forms of persistent activity that have been observed in certain brain areas. Transient inputs can flip a bistable cell or a bistable circuit back and forth between two states, but this does not resemble the examples of persistent activity that are continuously graded.