Mathematical theory predicts self-organized learning in real neurons

Free energy principle theory predicts how real neural networks self-organize during learning to distinguish between incoming information sources. Researchers tested the prediction using neurons from rat embryos grown in vitro on electrode grids. After training with mixed sensory patterns, the neurons became selectively responsive to one hidden input source over the other, matching learning expectations. The study also showed that changing neuron excitability with drugs before training can disrupt the learning process, supporting the model. Using early-session neural data, the researchers reverse engineered a predictive model that accurately forecast later neuronal responses and connection strength, with implications for understanding impaired learning and for modeling drug effects and psychiatric disorders.