Decoding Composition and Generalization of task representations in hierarchical task learning

Poster Session A - Saturday, April 13, 2024, 2:30 – 4:30 pm EDT, Sheraton Hall ABC

WooTek Lee¹ (woo-tek-lee@uiowa.edu), Jiefeng Jiang¹; ¹University of Iowa

Humans have remarkable abilities to learn and perform many tasks in ever-changing life. This ability is in part supported by the hierarchical organization of task representations. For example, the task of making Latte is composed of two simpler tasks of making espresso and steaming milk. Compositions sharing the same simple tasks further facilitate generalization to new compositions. For instance, knowing latte = espresso + milk and hot chocolate = milk + chocolate syrup help form a new task to combine espresso and chocolate to make mocha. Using electroencephalographic (EEG) data (n = 40), we tested the hypothesis that learning a complex task relies on reinstating the simple tasks involved. In the training stage, participants learned six simple tasks (coded as A-F) and then complex tasks consisting of two simple tasks (e.g., AB, BC). In the test staged, subjects performed new complex tasks that can be generalized from learned complex tasks (e.g., AC from AB and BC). We successfully replicated the behavioral generalization effect. In the EEG data, reinstatement was operationalized as decoding encompassed simple tasks from a complex task (e.g., A from AC, termed composition effect) and decoding the unpresented shared simple task (e.g., B from AC, termed generalization effect). Composition effect manifested during the training stage of complex task, while generalization effect became evident during the test stage. This finding was further supported by two correlations: (1) between behavioral and EEG generalization effects and (2) between EEG composition effect in training stage and EEG generalization effect in test stage.

Topic Area: EXECUTIVE PROCESSES: Goal maintenance & switching