ImDiffusion: Imputed Diffusion Models for Multivariate Time Series Anomaly Detection

TL;DR

ImDiffusion tackles multivariate time series anomaly detection by uniting time-series imputation with diffusion models, enabling robust modeling of complex temporal and inter-metric dependencies. The method employs grating masking and an unconditional diffusion process with ImTransformer, and leverages ensemble signals from multi-step denoising to boost accuracy and timeliness. Extensive offline benchmarks across six datasets show strong performance gains over 10 baselines, while a Microsoft production deployment demonstrates meaningful reliability improvements. Overall, the paper introduces a novel, production-proven paradigm that surpasses forecasting and reconstruction approaches in MTS anomaly detection by exploiting imputation-driven diffusion and ensemble reasoning.

Abstract

Anomaly detection in multivariate time series data is of paramount importance for ensuring the efficient operation of large-scale systems across diverse domains. However, accurately detecting anomalies in such data poses significant challenges. Existing approaches, including forecasting and reconstruction-based methods, struggle to address these challenges effectively. To overcome these limitations, we propose a novel anomaly detection framework named ImDiffusion, which combines time series imputation and diffusion models to achieve accurate and robust anomaly detection. The imputation-based approach employed by ImDiffusion leverages the information from neighboring values in the time series, enabling precise modeling of temporal and inter-correlated dependencies, reducing uncertainty in the data, thereby enhancing the robustness of the anomaly detection process. ImDiffusion further leverages diffusion models as time series imputers to accurately capturing complex dependencies. We leverage the step-by-step denoised outputs generated during the inference process to serve as valuable signals for anomaly prediction, resulting in improved accuracy and robustness of the detection process. We evaluate the performance of ImDiffusion via extensive experiments on benchmark datasets. The results demonstrate that our proposed framework significantly outperforms state-of-the-art approaches in terms of detection accuracy and timeliness. ImDiffusion is further integrated into the real production system in Microsoft and observe a remarkable 11.4% increase in detection F1 score compared to the legacy approach. To the best of our knowledge, ImDiffusion represents a pioneering approach that combines imputation-based techniques with time series anomaly detection, while introducing the novel use of diffusion models to the field.

Figures (9)

• Figure 1: Examples of reconstruction, forecasting and imputation modeling of time series for anomaly detection.
• Figure 2: Example cases of conditional/unconditional diffusion models for time series anomaly detection.
• Figure 3: An illustration of the the grating masking and the imputation process under this strategy.
• Figure 4: The training process of ImDiffusion.
• Figure 5: The ImTransformer architecture, with (a) the residual structure; and (b) the details of a residual block.