Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Unsupervised Machine Learning Identifies Unique Brain Structural Markers of Learning on a Gamified Working Memory Task in Older Adults

Poster Session C - Sunday, March 8, 2026, 5:00 – 7:00 pm PDT, Fairview/Kitsilano Ballroom

Dr. Chandramallika Basak¹ (), Soham Ghaisas², Yiyao Liu¹, Anirudhh Sowrirajan³, Kshitij Jhadhav³; ¹Center for Vital Longevity, ²Sardar Patel Institute of Technology, ³Indian Institute of Technology Bombay

Neurological markers of complex skill learning (> 10 hours) were assessed in 46 older adults (Mage = 71years) using the BirdWatchGame, a gamified working memory updating task. Individuals were clustered into two subgroups using unsupervised machine learning (ML) algorithms to five learning metrics:BlockSimpleScore, AsymptotePoint, RT for Correct Responses, RT Increase in Error-Correct Response Pairs, and Block d-prime. One-dimensional clustering for each metric and five-dimensional clustering combining all metrics revealed two distinct subgroups across all learning features. Statistical comparisons of 80 brain volumes and cortical thickness performed between clusters, correcting for multiple comparisons, showed significant differences in amygdala and basal ganglia for accuracy, but inferior frontal and frontal pole for RTand asymptote, and anterior cingulate for error monitoring. The 5-D clustering showed group differences in fronto-parietal hubs of cognitive control and amygdala. Our results suggest the importance of using unsupervised ML models in order to classify complex skill learning using both single and combined metrics, since the type of learning outcome assessed or emphasized during cognitive training can have implications on neurocognitive plasticity.

Topic Area: LONG-TERM MEMORY: Skill Learning