Mechanism Learning: Prototype-Anchored Mechanism Inference for Scientific Forecasting

2026-05-16

AI Summary

This work addresses the limitations of conventional direct state prediction methods in data-scarce, long-horizon, non-stationary, or high-dimensional complex settings. The authors propose a novel mechanism learning paradigm that predicts by inferring the locally active evolution mechanism at each step. Specifically, spatiotemporal segments are encoded into mechanism descriptors to construct a structured mechanism space, and prototype anchors are introduced to enable data-driven mechanism estimation. This approach effectively prevents collapse of the mechanism space while enhancing local consistency and prediction robustness. Extensive experiments on benchmarks including Burgers, WeatherBench2, and Lorenz96 demonstrate that the method significantly outperforms established baselines such as FNO, NODE, and LSTM, achieving state-of-the-art performance particularly under data scarcity and dynamic regime-switching conditions.

Abstract

Scientific forecasting typically relies on direct state prediction, an approach that grows brittle under data scarcity, extended horizons, non-stationary dynamics, or high-dimensional complexity. While raw state trajectories are highly sensitive in these regimes, underlying local evolution rules often exhibit robust reusability. We introduce mechanism learning, a framework that forecasts future states by estimating the currently active local mechanism. Our method compresses local spatiotemporal fragments into mechanism descriptors, forming a data-driven, structured mechanism space where proximity reflects similar local evolution rules. To ground these estimates in observed data, we utilize prototype anchors, a set of representative mechanisms that sparsely cover the space of local rules. We evaluate this approach on Burgers dynamics, WeatherBench2, and Lorenz96. Empirically, the learned mechanism spaces resist collapse and maintain strong local consistency. Compared to direct prediction and other models including FNO, NODE, LSTM, and reservoir-family methods, our framework demonstrates predictive gains in fragile regimes: it significantly improves switching stability in Burgers dynamics and achieves state-of-the-art performance both under the scarce-data fixed-horizon WeatherBench2 protocol and in intermediate-complexity Lorenz96. Ablation studies and drift diagnostics confirm that these improvements are driven by finite prototype anchoring rather than sheer latent capacity. Together, these results establish mechanism learning as a principled, robust alternative to direct state prediction in forecasting complex systems.

Problem

• scientific forecasting
• data scarcity
• non-stationary dynamics
• high-dimensional complexity
• direct state prediction

Innovation

• mechanism learning
• prototype anchors
• scientific forecasting
• local evolution rules
• structured mechanism space