Inductive bias anchors a scaling law for biological learning

Effective complexity multiplicatively sets biological learning speed, and a Bayesian scaling law predicts how quickly humans and mice improve from reward feedback based on prior–task alignment.

• Why it matters: Learning speed varies across individuals and task domains, and existing explanations lack a quantitative way to relate inductive bias alignment to measurable learning dynamics.
• What they did: A Bayesian learner model with iterative reward feedback yields a scaling law with two predictions: learning speed depends on effective complexity, and performance improves as a power law in experience; it is tested in a human motor learning dataset and a large-scale mouse learning dataset.
• The result: In humans, pre-learning behavioral measurements predict subsequent learning speed via effective complexity, and in mice, a two-parameter rescaling collapses learning curves onto a universal function consistent with the theory.

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