MADLLM: Multivariate Anomaly Detection via Pre-trained LLMs
This addresses the problem of effective anomaly detection in multivariate time series for domains like monitoring systems, though it is incremental as it builds on existing LLM and embedding methods.
The paper tackles the misalignment between multivariate time series data and text modalities in pre-trained LLMs for anomaly detection by introducing MADLLM, which uses a triple encoding technique to integrate patch, skip, and feature embeddings, resulting in outperforming state-of-the-art methods on public datasets.
When applying pre-trained large language models (LLMs) to address anomaly detection tasks, the multivariate time series (MTS) modality of anomaly detection does not align with the text modality of LLMs. Existing methods simply transform the MTS data into multiple univariate time series sequences, which can cause many problems. This paper introduces MADLLM, a novel multivariate anomaly detection method via pre-trained LLMs. We design a new triple encoding technique to align the MTS modality with the text modality of LLMs. Specifically, this technique integrates the traditional patch embedding method with two novel embedding approaches: Skip Embedding, which alters the order of patch processing in traditional methods to help LLMs retain knowledge of previous features, and Feature Embedding, which leverages contrastive learning to allow the model to better understand the correlations between different features. Experimental results demonstrate that our method outperforms state-of-the-art methods in various public anomaly detection datasets.