Time-Prompt: Integrated Heterogeneous Prompts for Unlocking LLMs in Time Series Forecasting
Researchers have developed new frameworks to enhance the use of large language models (LLMs) for time series forecasting. PaP-NF utilizes a Prefix-as-Prompt mechanism to align time series data with a frozen LLM, enabling probabilistic forecasting and uncertainty quantification. Time-Prompt integrates learnable soft prompts and textualized hard prompts to guide LLMs, fusing temporal and textual data for improved forecasting accuracy. MAP4TS, though withdrawn, proposed a multi-aspect prompting framework incorporating classical time-series analysis into prompt design for better LLM performance. AI
IMPACT New LLM-based frameworks offer improved accuracy and uncertainty quantification for time series forecasting tasks.