Kolmogorov Arnold Networks and Symbolic Regression can recover dynamics from time series data

Modeling with systems of differential equations requires prior knowledge to create a fully specified model reflecting our understanding of biological systems, but sometimes we don’t have a complete understanding of the systems we are interested in. If we have time series data of our variables of interest, multilayer perceptron models can take the place of unknown terms in our equations to produce solutions that fit the data well but due to the nature of multilayer perceptron models are not very interpretable. Interpretability can be improved by combining the simpler compositionality of Kolmogorov Arnold Networks with symbolic regression, allowing for the discovery of unknown terms from time series data. In this poster, we leverage the interpretability of Kolmogorov Arnold Networks with symbolic regression to recover a logistic growth term from time series data generated by a predator-prey model.

Kolmogorov Arnold Networks and Symbolic Regression can recover dynamics from time series data

Monday, July 14 from 6:00pm - 9:00pm in

SMB2025 Follow

Lucas MacQuarrie

"Kolmogorov Arnold Networks and Symbolic Regression can recover dynamics from time series data"

#SMB2025 Follow