FlowMixer: A Constrained Neural Architecture for Interpretable Spatiotemporal Forecasting
This addresses the need for interpretable forecasting in domains like physics and engineering, though it appears incremental by combining existing concepts like non-negative matrix mixing and reversible mappings.
The paper tackled the problem of interpretable spatiotemporal forecasting by introducing FlowMixer, a neural architecture that uses constrained matrix operations to model structured patterns, resulting in robust long-horizon forecasting capabilities across diverse domains, including chaotic attractors and turbulent flows.
We introduce FlowMixer, a neural architecture that leverages constrained matrix operations to model structured spatiotemporal patterns. At its core, FlowMixer incorporates non-negative matrix mixing layers within a reversible mapping framework-applying transforms before mixing and their inverses afterward. This shape-preserving design enables a Kronecker-Koopman eigenmode framework that bridges statistical learning with dynamical systems theory, providing interpretable spatiotemporal patterns and facilitating direct algebraic manipulation of prediction horizons without retraining. Extensive experiments across diverse domains demonstrate FlowMixer's robust long-horizon forecasting capabilities while effectively modeling physical phenomena such as chaotic attractors and turbulent flows. These results suggest that architectural constraints can simultaneously enhance predictive performance and mathematical interpretability in neural forecasting systems.