In December, our PIs, Carsten Dormann and Joschka Boedecker, published a paper in the Ecology Letters journal on their work on process-informed neural networks (PINNs).
PINNs are neural networks that incorporate process based models into their prediction process. This combination is particularly useful in the field of ecology, as collected data is often sparse and incomplete. Under these circumstances, deep learning methods alone become prone to finding correlations that may not have the right causality, impacting their predictive outcomes. Process based models, on the other hand, derive from the established scientific understanding of the field and are not constrained by data sparsity.
By combining both, the authors of this work have shown that PINNs outperform the prediction potential of process based models and neural networks alone in data sparse regimes.
You can find the open access publication here.
Institute of Medical Biometry and Statistics,
Faculty of Medicine and Medical Center –
University of Freiburg
