Multi-Modal Foundation Models For Alzheimer’s Detection

Summary: Our project leverages multimodal foundation models to advance the detection and understanding of Alzheimer’s disease. By integrating diverse data modalities—including MRI imaging, clinical data, and genetic information—we aim to create a unified, scalable framework capable of capturing complex patterns associated with the disease. The foundation model approach allows us to build a shared representation across modalities, enhancing predictive accuracy and interpretability. This ensures that our models not only excel at diagnosis but also provide insights into contributing factors, potentially aiding early intervention and personalized treatment strategies.

Leveraging Uncertainty Quantification to Improve Forecasting

Summary: This project explores the integration of uncertainty quantification (UQ) into forecasting systems, inspired by the principles of predictive coding. Predictive coding, a framework rooted in neuroscience, emphasizes the brain’s ability to minimize prediction errors by dynamically adjusting its internal models. We aim to mimic this adaptive mechanism in computational forecasting via different UQ techniques, including Monte Carlo Dropout, Bayesian Networks, and Calibrated Regression. This approach enhances the model’s robustness and adaptability, particularly in noisy and chaotic short-term forecasting regime.