Brain Data Value Quantified for Machine Learning Training

A new theoretical framework quantifies the value of neural recordings for training machine learning models. Researchers derived scaling laws showing how model performance improves with brain data versus task labels. Using a linear Gaussian model, they calculated exchange rates between brain samples and task samples, dependent on task-brain alignment, noise, and latent dimension. The work addresses when and how much benefit neural data provides, offering a mathematical foundation for NeuroAI. The study was published on arXiv under ID 2605.09243.