Modeling Primacy, Recency, and Cued Recall in Serial Memory Task Using On-Center Off-Surround Recurrent Neural Network

The serial recall paradigm has long been used to study short-term memory. Previous experiments have consistently revealed two key phenomena: the primacy effect and the recency effect. Essentially, it is easier to recall items at the beginning and end of a sequence compared to those in the middle. In this study, we present a single-layer fully connected recurrent neural network with self-excitation and mutual inhibition. By providing transient input to the network, we observed a dynamic steady-state output. We examined this output pattern to determine which inputs were ultimately “remembered” at the end of the simulation. Our results demonstrate that this network can replicate the serial recall curve observed in empirical studies when higher inhibition values are used. Additionally, it can account for the capacity limitation commonly observed in serial recall tasks. By varying the presentation duration in the model, we successfully explain both the primacy and recency effects within the same network. Furthermore, when we introduced cues to a single item in the sequence by elevating its input, we observed a decrease in the recall probability of neighboring items. In summary, our findings suggest that the dynamics of a single-layer recurrent on-center off-surround neural network can provide insights into the mechanisms underlying primacy, recency, and cued recall effects.