Computational Memory Lab

Welcome to the Computational Memory Lab. We study human verbal memory behaviour and its basis in cognitive and neural processes, using mathematical modelling, behavioural experiments, and brain imaging (EEG and fMRI).

Visit our Research page to try three hands-on interactive demos that let you experience our key findings firsthand!

Explore Our Interactive Research Demos →

At the Computational Memory Lab, we study how the brain encodes and retrieves relational memory — not just what items were studied, but the relationships between them. The basic building blocks of relational memory are associations (e.g., learning that garlic–vampires go together) and order (sequences like phone numbers or song lyrics). We combine mathematical models, behavioural experiments, and brain-activity measures (EEG and fMRI) to systematically probe the cognitive and neural basis of how humans remember relational information.

One major focus is relational influence: when you study a pair of words together, one item can shape which features of the other get encoded. For example, RIVER and VAULT each cause you to attend to very different aspects of the word BANK. We have developed mathematical models showing that this kind of meaning-driven influence can explain why associative recognition memory is sometimes preserved in amnesia, and we run behavioural experiments to measure just how much relational influence shapes memory in healthy participants.

A second focus is order within associations — whether people remember which item came first in a pair (was it WALL BABY or BABY WALL?). We have found new evidence in favour of associative chaining models, in which order is stored within the association itself, rebutting arguments that were long thought to rule chaining out. At the same time, neither chaining nor positional-coding models alone can explain all of the data, and developing a compact hybrid theory that handles both is a major ongoing goal. We also study associative interference — what happens when associations compete because they share a common element — and have identified novel neural signatures of both the interference itself and its resolution in the brain.

To connect brain activity to behaviour and models, we use EEG to record signals during memory tasks. We have shown that theta oscillations (4–8 Hz rhythmic activity) are robustly related to successful encoding of associations, and we apply classifier-based machine-learning approaches to identify distributed patterns of brain activity that predict later memory success. To detect oscillations reliably in the first place, Dr. Caplan co-developed BOSC (Better OSCillation Detection), a method now used by over 20 research groups worldwide and applied across species and recording modalities.

The lab is part of the Alberta Cognitive Neuroscience Group and the Neuroscience and Mental Health Institute. Our EEG equipment is funded by the Canada Foundation for Innovation (CFI) and the Natural Sciences and Engineering Research Council of Canada (NSERC), which also provides our operating funding.