Problem
Research questions and friction points this paper is trying to address.
Large Language Models
Multivariate Time Series
Structural Patterns
Seasonal and Trend Components
Frequency-Semantic Alignment
FreSia: Frequency-semantic instantiation and alignment for multivariate time series analysis
🤖 AI Summary
This study addresses the inherent difficulty of large language models in perceiving seasonal and trend structures within time series from discrete numerical values. To this end, this work proposes a frequency-aware framework that achieves effective alignment between semantic and temporal-frequency spaces through frequency-domain structural distillation and global context learning. The core innovations include a frequency-guided prompting mechanism and a global-driven context learning component, designed to bridge the time-frequency domain gap and integrate multimodal information. By synergizing frequency analysis, large language models, prompt engineering, and a global CLS probe, the proposed method demonstrates substantial improvements across eight benchmarks, reducing mean squared error and mean absolute error by an average of 13.48% and 8.06%, respectively.
📝 Abstract
Large Language Models (LLMs) have shown strong potential in multivariate time series forecasting and anomaly detection. Existing studies predominantly inject temporal information into LLMs via direct numerical tokenization or heuristic textual descriptions. However, LLMs still face difficulty in perceiving the underlying structural patterns of numerical time series, particularly the seasonal and trend components obscured by discrete numerical tokens. To bridge this gap, we propose FreSia, a frequency-aware framework that establishes an effective alignment between the semantic space of LLMs and the frequency space of time series. Specifically, FGPrompt, a Frequency-Guided Prompt mechanism within FreSia, distills the frequency-domain structures of time series and projects them into prompts tailored to the semantic space of LLMs. Furthermore, we introduce a Global-driven Context Learning (GCL) component, which uses a global CLS-driven probe to generate global context to bridge the time-frequency domain gap and fuse the multi-modal information. Experiments on eight forecasting benchmarks show that FreSia achieves average improvements of 13.48% and 8.06% in MSE and MAE, respectively.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Frequency-Semantic Alignment
Large Language Models
Multivariate Time Series
Frequency-Guided Prompt
Global-driven Context Learning
🔎 Similar Papers
No similar papers found.
