Date of Graduation
Doctor of Philosophy (PhD)
Dr. Anthony Wright
Dr. Wei Ji Ma
Dr. Anne Sereno
Dr. Valentin Dragoi
Dr. James Pomerantz
Visual working memory (VWM) is the temporary retention of visual information and a key component of cognitive processing. The classical paradigm for studying VWM and its encoding limitations has been change detection. Early work focused on how many items could be stored in VWM, leading to the popular theory that humans could remember no more than 4±1 items. More recently, proposals have suggested that VWM is a noisy, continuous resource distributed across virtually all items in the visual field, resulting in diminished memory quality rather than limited quantity. This debate about the nature of VWM has predominantly been studied with humans. Nevertheless, nonhuman species could add a great deal to the debate by providing evidence related to evolutionary continuity (similarities and/or differences) and model systems for investigating the neural basis of VWM. To this end, in the first aim, we tested monkeys and humans in virtually identical change detection tasks, where the subjects identified which memory item had changed between two displays. In addition to the typical manipulation of the number of items to-be-remembered (2-5 oriented bars), we varied the change magnitude (degree of orientation change) – a critical manipulation for discriminating among leading models of VWM encoding limitations. We found that in both species VWM performance was best accounted for by a model in which memory items are encoded in a noisy manner, where quality of memory is variable and on average decreases with increasing set size.
The second aim focused on the decision-making component of change detection, where observers use noisy sensory information to make a judgment about where the change occurred. We tested monkeys and humans in the same change detection task (Aim 1), but with ellipses that varied in their height-to-width ratio so that their reliability of communicating orientation discrimination could be manipulated. The high-reliability ellipses were long and narrow, and the low-reliability ellipses were short and wide. We compared models that differed with respect to how the observers incorporate knowledge of stimulus reliability during decision-making. We found that in both species performance was best accounted for by a Bayesian model in which observers take into account the uncertainty of sensory observations when making perceptual judgments, giving more weight to more reliable evidence.
The comparative results across these related primate species are suggestive of evolutionary continuity of basic VWM processing in primates generally. These findings provide a strong theoretical foundation for how VWM processes work and establish rhesus monkeys as a good animal model system for physiological investigations to elucidate the neural substrates of VWM processing.
visual working memory, change detection, memory limitations, perception, Bayesian modeling, decision making