Math Encoding In The Human Brain
Ecole Normale Supérieure, Paris
I am analyzing a large dataset of intracranial electrophysiological recordings in humans, who performed a variety of mathematical cognitive tasks, from basic number identification to mental calculation. The recordings were performed on patients with refractory epilepsy during their presurgical evaluation at Stanford University Medical Center. The database includes 100 subjects implanted with grids of electrodes or with penetrating depth electrode arrays. In total, we have ~60h of task related recordings from 9,000 sites across all patients, covering roughly the whole brain. This constitute the largest and more comprehensive database of intracranial recordings in humans performing math. This project has 5 main goals: 1) identify a whole-brain network of regions selectively engaged in mathematical processing, 2) characterize their precise role in the calculation processing chain; 3) investigate their temporal and oscillatory dynamics; 3) probe at the single subject level how these regions relate to the canonical intrinsic brain networks derived from resting-state functional connectivity; 5) test their causal relationship with behavioral performance. For goals 1 and 2, I am using a predictive statistical framework with machine learning algorithms to fit encoding and decoding models on the dataset. Encoding models aim at predicting brain activity from a series of stimuli features (e.g. from low level visual parameters of the images to more high level abstract dimensions such as numerosity and operation type). Complementary decoding models aim at predicting the series of stimuli features from brain activity.