Presenter: Ernesto Suárez, Ph.D.

Accurately estimating entropy in high-dimensional, heavily correlated systems remains a fundamental challenge in statistical mechanics and information theory. In the extreme sparse regime, where the system’s state space K vastly exceeds the number of observations (K >> N), traditional frequency-based estimators fail catastrophically. To address this, we introduce a methodology that leverages a mutual information-based metric space and the sum of conditional entropies to infer the unobserved state space. We demonstrate that this approach successfully recovers the entropy of complex, high-dimensional systems even when empirical samples are unique (no repeats), and we showcase its broad utility across applications ranging from predicting the free-energy difference of biomolecules to studying high-dimensional single-cell transcriptomic states. Attendees should have a basic knowledge of statistics.

This will be a hybrid event. Please register at this link.

This session will be recorded, and materials will be shared with attendees a few days after the event.

For additional details and questions, please contact Natasha Pacheco (natasha.pacheco@nih.gov), Advanced Biomedical Computational Science group, Frederick National Laboratory for Cancer Research.