Matthew Costello has been studying how aging affects cognition and perception for close to 10 years. But answers to the questions of exactly how and why visual working memory declines in older adults have still eluded him and other researchers. Now, he is taking an information processing approach to this topic that affects so many people and will eventually affect us all.
Visual working memory is key to how we navigate in the world, how we see and remember things, and countless other everyday tasks. “When we walk down the grocery store aisle, we take in visual information as a continuous stream, temporarily store that information, and then modify our actions based upon those inputs,” says Costello, a developmental psychologist at the University of Hartford. But as people age, the relative weighting of information and the efficiency of processing that information stream changes, leading to increased forgetting, such as where someone parked the car. While such minor forgetting is expected in normal, healthy older adults, one of the major challenges in cognitive aging research, Costello says, is determining the lines between normal and abnormal.
Publishing in the Journal of Cognitive Neuroscience, Costello and colleague Aaron Buss at the University of Tennessee have developed a dynamic neural field (DNF) model to look more closely at age-related differences in visual working memory. Biologically-grounded, the model consists of three dynamically interactive layers of simulated neurons that encode visual stimuli, maintain representations of those stimuli in an active state, and compare the contents of memory to visual stimuli in order to make decisions about differences or similarities.
CNS spoke with Costello about this new study exploring visual working memory in older adults and about future work to apply this work to real-world settings.
CNS: What have we known previously about visual working memory in older adults compared to other age groups?
Costello: A consistent finding is that older adults have degraded working memory relative to younger adults. This is not considered pathological, but a normal, albeit unfortunate, expression of the aging brain. What researchers are now debating about is how to conceptualize such declines. Some have argued that aging results in a reduced cap on the number of items that can be stored, whereas others have argued that the question is not so much the cap but the quality of the information stored, with older adults exhibiting degraded quality of visual working memory. Because our specific approach to visual working memory has been informed from dynamic systems theory, we tend to approach visual working memory as just one component within a larger connected system. By our account, age-related declines in visual working memory reflect underlying changes to other cognitive areas, such as inhibition levels and perceptual intake.
The beauty of computational modeling is that it offers a theoretically-driven mechanistic system to unify the cognitive and the biological.
CNS: What was the new insight you were seeking?
Costello: The major goal of our study was to apply a computational model to simulate the age group differences in a visual working memory task. Participants were tested in a change detection task. They viewed an array of items on the computer screen, which after a brief delay, was replaced with a similarly arrayed screen. On half the trials, there was a change – one item was altered in either its shape or color. Participants responded to whether the second display was the same or different from the first display. We conducted this experiment with both younger and older adults and found an age-related decline in task performance. Further, there was a characteristic response pattern for the older adults of a ‘same bias’, in which they responded more readily and accurately to ‘same’ trials. We then tried to replicate these age group differences using a DNF model.
CNS: Tell us a bit more about DNF models.
Costello: DNF models have been used over the past 20 years for modeling neural activity in the context of visual working memory, although it has generally been applied to children and younger adult populations. The central novelty to our study was that we were applying the DNF model to capture the older adult performance, which had never been done before in the context of visual working memory. Our study was broadly successful in simulating the age group performance patterns using modifications to the young model parameters. This opens the door for new insights into how and why age-related differences arise, a move beyond merely identifying cognitive mechanisms subject to age effects, or even brain regions associated with such deficits. The beauty of computational modeling is that it offers a theoretically-driven mechanistic system to unify the cognitive and the biological.
CNS: What were you most excited to find?
Costello: We were surprised at how closely the old and young models mirrored the participant behavioral patterns based upon subtle adjustments to the model parameters. We manipulated various parameters of the model to explore which modifications would result in model behavior that replicated performance of older adults. It turns out that a subset of parameters worked in this regard. Models that replicated older adult performance had either wider inhibitory projection, wider inhibitory and excitatory projections, or weaker input to inhibitory populations. These findings are consistent with the literature insofar as inhibitory control is a prominent theory in the literature, although note that here we refer to inhibition within the context of neural responses. We were also excited to find that increasing stochastic noise within the model did not help the model-behavioral fit. This is important because one prominent theory of cognitive aging is that older adult performance is characterized by increased neural noise. Although our results don’t necessarily eliminate this possibility, they certainly don’t support it.
CNS: What do you most want people to understand about this work?
Costello: The central takeaway is that DNF models can be used as a powerful theoretical tool to ground our understanding of cognitive function, and changes in cognitive function over time, in neural processes. While the DNF model is relatively new to the gerontological literature, it has been successful employed in visual working memory studies with other age groups, and now we have successfully applied it to the older adult population. This could lead to future insights into how aging can alter the relationship between cognitive subsystems and their underlying neural architecture.
CNS: What’s next for this line of work?
Costello: My collaborator on this paper Aaron Buss and I, along with colleagues at the University of Connecticut, currently have an NIH grant application under consideration that will allow us to extend this work into several different areas. First, we want to continue this existing line of visual working memory work using fMRI, which would allow us to integrate the ‘big three’ – behavioral performance, computational modeling, and neuroimaging. The model framework we have been developing is capable of simulating both real-time behavioral responses as well as real-time hemodynamic responses, a unique achievement for computational theories of cognition. We also want to direct our studies into a very concrete application of visual working memory for the elderly – driving a car. We hope to run older adults in a driving simulator and explore how selective attention and visual working memory factor into older adult driving performance, and to understand better the underlying cortical activations responsible for their driving performance.
-Lisa M.P. Munoz
