What is it about?

It is challenging to conceptualize how memory works in the brain. Neural networks can be used to simulate complex memory processes and to help conceptualize how memory works. We used neural network models to help figure out how a “distributed” memory system might work, that is, how memory might work in a way that does not separate memory functions into discrete categories but rather views memory as one continuous, distributed process supported by multiple brain networks. Critically, we also had to modify traditional neural network architecture to get around the widespread problem of catastrophic interference that prevents neural networks from retaining information learned. With catastrophic interference, learning new information overwrites previously learned information. By providing neural networks with a way to distinguish new information during learning, like how context in our everyday lives helps us learn and remember new information, we can help neural networks remember. We show that by viewing memory functioning through this alternative lens, we can have different expectations about how memory breaks down in Alzheimer’s disease. We used this new way of thinking about memory to help design a digital memory test designed to be sensitive to early memory changes that occur in Alzheimer’s disease.

Featured Image

Why is it important?

There are several unique aspects of this work: 1. We demonstrate how neural networks can be used as process models to help us think through complex cognitive processes. 2. We show how learning and remembering can occur without separate “short-term” and “long-term” memory stores. 3. We modified traditional neural network architecture to prevent catastrophic interference. This may have important implications for building artificial intelligence networks that can learn more efficiently. 4. We apply these new conceptual insights to create a new digital memory test that is designed to be sensitive to changes in memory that occur in the early stages of Alzheimer’s disease.

Perspectives

Forgetting is a cardinal feature of Alzheimer’s disease and is easy to observe once an individual reaches the mild to moderate stage of dementia. Accordingly, neuropsychologists often emphasize the ability to remember information after a delay when using clinical neuropsychological tests to help inform if an individual may have Alzheimer’s dementia. However, memory decline is a gradual process and learning is required before any forgetting can occur. As the field focuses more on early detection, an over-emphasis on “forgetting” may prevent us from considering alternative testing paradigms better suited for digital cognitive testing approaches that have the potential to reach more people. This article provides a way to understand how learning and forgetting are intertwined and offers a different view about what underlying mechanisms may contribute to learning and memory decline in Alzheimer’s disease. Specifically, neural network models suggest that a breakdown in context, or the ability to organize information as it is learned, may play a key role. New tests that consider this conceptualization may help with earlier detection of subtle memory decline and ultimately aid early prevention efforts.

Nikki Stricker
Mayo Clinic College of Medicine and Science

Read the Original

This page is a summary of: Neural network process simulations support a distributed memory system and aid design of a novel computer adaptive digital memory test for preclinical and prodromal Alzheimer’s disease., Neuropsychology, August 2022, American Psychological Association (APA),
DOI: 10.1037/neu0000847.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page