Predicting the Unpredictable: LLM-powered Long-term Chaotic Time Series Forecasting under Short-term Observations
PAC-LLM adds phase-space features to an LLM forecaster for chaotic systems with only short observation windows.
The paper says standard approaches struggle when there is not enough trajectory history to learn long-run dynamics. Its framework fuses learned phase-space features with textual information, using an auxiliary feature module and gated weighting for multivariate coupling. The authors report stronger short- and long-term results than fine-tuned and zero-shot baselines on representative chaotic systems. ArXiv · AI/CL/LG's note
The paper says standard approaches struggle when there is not enough trajectory history to learn long-run dynamics. Its framework fuses learned phase-space features with textual information, using an auxiliary feature module and gated weighting for multivariate coupling. The authors report stronger short- and long-term results than fine-tuned and zero-shot baselines on representative chaotic systems. ArXiv · AI/CL/LG's note
score 3