Same Movements, Different Skills: How Context Shapes Learning in Humans and Neural Networks

Context determines whether identical finger-sequence practice is integrated or split into different motor skills, with splitting strongest when strong contextual inputs align with a high initial weight scale in recurrent networks.

• Why it matters: The work addresses how experience becomes shared versus separated representations, testing whether identical movement sequences should generalize across tasks when only context changes.
• What they did: Simple recurrent networks were trained on the same movement sequence across contexts, varying contextual input strength and initial weight scale; humans practiced the same finger sequence while alternating two performance-irrelevant contexts.
• The result: Splitting in networks was greatest when both strong contextual inputs and high initial weight scale coincided, and humans showed context-specific learning curves; state-space modeling supported split learning states in humans and weak coupling in the most context-dependent networks.

AI summary of the abstract · may contain errors · not medical advice · read the paper.