Smart segmentation supports transfer learning

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

Anna Jafarpour¹ (annaja@uw.edu), Robert Knight^2,3, Elizabeth Buffalo^1,4; ¹University of Washington, ²UC Berkeley, ³Helen Wills Neuroscience Institute, ⁴National Primate Center

We are faster in making sense of and learning new challenges when we have accomplished a similar challenge before. Speeded learning based on previous experience, transfer learning, is thought to depend upon the identification of a stable task structure. However, many common tasks contain a hierarchical or nested stable structure, in which subtasks vary in behavioral relevance, and the extent to which transfer learning mechanisms represent nested stable task structures is currently unknown. We tested the learning and memory behavior of healthy adults and examined learning of a fixed sequence of tasks where some subtasks acted as distractors and other subtasks determined the outcome of successful navigation to a goal. The relationship between subsequent memory of subtasks and reaction time during learning, together with computational modeling of learning and segmentation, revealed that participants who demonstrated transfer learning adopted a smart task segmentation strategy including the separation of distracting subtasks. This research was supported by the National Institute of Health, National Institute of Mental Health, K99MH120048-01 (A.J.), the National Institute of Neurological Disorders and Stroke NS21135 (RTK), and Brain Initiative, 1U19NS107609 (E.A.B.).

Topic Area: LONG-TERM MEMORY: Skill Learning