Search papers, labs, and topics across Lattice.
The paper introduces Spatiotemporal Forecasting as Planning (SFP), a model-based reinforcement learning framework that addresses stochasticity and non-differentiable metrics in spatiotemporal forecasting. SFP uses a generative world model to simulate future states and employs a beam search-based planning algorithm, guided by non-differentiable domain metrics, to identify high-reward future sequences. These sequences are then used as pseudo-labels to iteratively self-train a base forecasting model, leading to improved prediction accuracy and performance on metrics like extreme event capture.
Model-based RL can leverage non-differentiable domain metrics to drastically improve spatiotemporal forecasting, especially for capturing extreme events.
To address the dual challenges of inherent stochasticity and non-differentiable metrics in physical spatiotemporal forecasting, we propose Spatiotemporal Forecasting as Planning (SFP), a new paradigm grounded in Model-Based Reinforcement Learning. SFP constructs a novel Generative World Model to simulate diverse, high-fidelity future states, enabling an"imagination-based"environmental simulation. Within this framework, a base forecasting model acts as an agent, guided by a beam search-based planning algorithm that leverages non-differentiable domain metrics as reward signals to explore high-return future sequences. These identified high-reward candidates then serve as pseudo-labels to continuously optimize the agent's policy through iterative self-training, significantly reducing prediction error and demonstrating exceptional performance on critical domain metrics like capturing extreme events.
