The School of Science invites members from each of its 5 different departments to share the research they’ve been doing at TCNJ. This also gives faculty members a chance to share the outstanding contributions TCNJ students have made when participating in undergraduate research. Dr. Nardini was asked to share his research!
Title: Mechanistic machine learning to predict biological motion
Abstract: Data scientists may view mechanistic math modeling and machine learning as entirely separate entities. Modelers incorporate domain expertise into mechanistic models to understand and predict sparsely-sampled experimental data. In machine learning, on the other hand, one extracts patterns from large datasets to develop highly predictive computational algorithms. In this talk, I will present recent research in mechanistic machine learning, where modelers harness the advantages of both approaches by developing machine learning methods with mechanistic insights built into the data training process. This leads to interpretable machine learning algorithms that can accurately predict new data not seen during model training. In particular, I will highlight the use of a novel methodology to incorporate biological knowledge into the neural network training procedure as a means to predict simulated biological data of collective migration. Here, mechanistic machine learning allows us to make predictions that are not possible when using either mechanistic modeling or machine learning alone. This simulated biological data is motivated by experiments in cell biology, but is also relevant to microbiology, toxicology, and ecology.