GlucoFM: A Dual-Stream Foundation Model for Continuous Glucose Monitoring

Google Research and UNSW Sydney have introduced GlucoFM, a compact foundation model that processes continuous glucose monitoring data by separating long-term physiological trends from immediate, transient events. By decoupling these data streams, the 0.72M-parameter model improves performance on metabolic monitoring tasks compared to traditional methods that treat glucose readings as a single sequence. This development offers a more efficient way to interpret dense health data, potentially enhancing the accuracy of automated metabolic tracking and diagnostics.