Interference between similar memories increases the dimensionality and dispersion of recalled content

Poster Session C - Sunday, April 14, 2024, 5:00 – 7:00 pm EDT, Sheraton Hall ABC

Julian Gamez¹, Anisha Babu², James Murray³, Brice Kuhl⁴; ¹University of Oregon

Although countless memory studies have documented that similarity between memories produces interference and forgetting, relatively less is known about how memories adapt to interference so that they are successfully recalled. In rodents, high-dimensional neural representations have been shown to be critical for flexible behavior and cognition (e.g., Rigotti et al., 2013). Here, we tested whether memories adapt to interference through increases in the dimensionality of memory content. Human subjects (N = 120) extensively learned 6 face-scene associations. For half of the subjects, all of the scenes were from the same visual category and highly similar (e.g., 6 ‘libraries’; competitive condition); for the other half of subjects, each scene was from a distinct visual category (non-competitive condition). After learning, participants completed a recall task in which they were shown each face and typed a description of the corresponding scene. To quantify memory dimensionality, Natural Language Processing was applied to the scene descriptions, yielding a semantic embedding for each scene (memory). Principal Component Analysis applied to these embeddings revealed that, compared to the non-competitive condition, participants in the competitive condition contributed relatively less to early components (1-5) and relatively more to later components (6-20), suggesting interference resulted in higher-dimensional memory representations. Additionally, K-means clustering (with K=6 categories) applied to the embeddings revealed that the distance to cluster centroids (dispersion) was significantly greater in the competitive condition compared to the non-competitive condition. Collectively, these findings support the idea that similarity between memories induces adaptive increases in the dimensionality of memory content.

Topic Area: LONG-TERM MEMORY: Episodic