LLMODE Framework for Spatio-Temporal Forecasting
LLMODE: Aligning ODEs with LLMs via Gated Token Injection for Irregular Spatio-Temporal Forecasting
LLMODE handles irregular spatio-temporal data more effectively than standard approaches, making it valuable for urban planning and environmental monitoring applications in ASEAN regions with irregular data collection.
LLMODE aligns ODEs with LLMs via gated token injection for irregular spatio-temporal forecasting. The framework uses a graph-aware ODE encoder to reconstruct irregular graph observations as a continuous-time latent trajectory. A Fixed-Budget Perceiver Resampler compresses variable-length trajectories into dynamic memory tokens while statistical descriptors become context memory tokens. Dual-source gated cross-attention injects both memories into frozen LLMs, achieving competitive performance on three urban datasets and two physical-dynamics benchmarks.
LLMODE: Aligning ODEs with LLMs via Gated Token Injection for Irregular Spatio-Temporal Forecasting
Large language models (LLMs) have shown promise for spatio-temporal forecasting, but existing approaches often rely on regularly sampled token sequences and struggle with irregular observations because of temporal asynchrony, representation-space misalignment, and limited context windows. We propose LLMODE, a token-efficient framework for irregular spatio-temporal forecasting with a frozen LLM backbone. LLMODE first uses a graph-aware ODE encoder to reconstruct irregular graph observations as a continuous-time latent trajectory. A Fixed-Budget Perceiver Resampler then compresses this variable-length trajectory into a fixed number of dynamic memory tokens. In parallel, compact statistical descriptors are encoded and resampled into context memory tokens. A dual-source gated cross-attention module injects both memories into the frozen LLM, enabling controlled utilization of external spatio-temporal evidence. Experiments on three real-world urban datasets and two physical-dynamics benchmarks show competitive overall performance, with clearer advantages under sparse or dynamically complex irregular sampling. Additional evaluations on unseen urban regions further demonstrate strong zero-shot generalization without adaptation.
