FreSia Aligns LLM Semantic Space with Time Series Frequency Domain
October 6, 2026
The FreSia framework improves multivariate time series analysis by projecting frequency-domain structures into LLM-compatible prompts via FGPrompt. This mechanism helps LLMs perceive seasonal and trend components that are typically lost during standard numerical tokenization.
HOW THIS AFFECTS YOU
●
builderThis approach offers a way to enhance the utility of LLMs for specialized numerical forecasting tasks.
●
researcherYou can improve time series forecasting accuracy by aligning frequency-domain data with LLM semantic spaces.