UIUC Quant Brownbag

Memory research has revealed two conflicting phenomena. Computational and behavioral work on eyewitness memory suggests that correlations between memory signals enhance memory, whereas research on visual memory for simple features, such as color, indicates that similarity between memory signals hurts memory. In the current work, we propose a simple integrative framework that reconciles these conflicting effects of correlated noise and similarity. Using a simple signal detection model, we show formally that memory performance will always be optimized when similarity between memory signals is minimized, but only if specific ordinal constraints are met. We examine the generality of this framework across a range of memory tasks and stimuli. Together, we outline basic principles that can be used to predict when similar items being present at test may hurt or facilitate detection and discrimination performance in memory tasks. We also discuss how these principles can be used in the development of ecological models of memory that formally incorporate the latent similarity structure of stimuli spaces.