Robust maintenance of both stimulus location and amplitude in a working memory model based on dendritic bistability

Working memory is a core feature of cognition that enables items to be maintained and manipulated over short durations of time. Stored information can be binary, such as the presence or absence of an object, or graded, such as the graded intensity or location of a feature. Current computational models of working memory cannot robustly maintain both the graded intensity and spatial location of a stored item. Here, we show how this limitation can be overcome if neurons contain multiple bistable dendritic compartments. First, we illustrate the core mechanism for the storage of graded amplitude information in a simple spiking "autapse" circuit consisting of a single neuron connected to itself. Second, we reduce this model to a rate-based model that permits analytic understanding. Third, we implement this mechanism within a spatially extended architecture in which the spatial location of an item is encoded by the set of active neurons. In contrast to classic spatial working memory models, which only encode the binary presence of an item at a given location, the multi-dendrite-neuron model robustly encodes both the amplitude and location of an item in working memory in a noise-resistant manner and without requiring fine tuning of parameters. We show analytically that the key mechanism permitting the storage of amplitude information is equivalent to that of the simpler autapse circuit. This work provides a solution to the problem of encoding graded information in spatial working memory and demonstrates how dendritic computation can increase the representational capacity and robustness of working memory.