TAIL-MIL: Time-Aware and Instance-Learnable Multiple Instance Learning for Multivariate Time Series Anomaly Detection (AAAI 2025)

The source code for the paper "TAIL-MIL: Time-Aware and Instance-Learnable Multiple Instance Learning for Multivariate Time Series Anomaly Detection", which was presented at AAAI 2025.

Abstract

This study addresses the challenge of detecting anomalies in multivariate time series data. Considering a bag (e.g., multi-sensor data) consisting of two-dimensional spaces of time points and multivariate instances (e.g., individual sensors), we aim to detect anomalies at both the bag and instance level with a unified model. To circumvent the practical difficulties of labeling at the instance level in such spaces, we adopt a multiple instance learning (MIL)-based approach, which enables learning at both the bag- and instance- levels using only the bag-level labels. In this study, we introduce time-aware and instance-learnable MIL (simply, TAIL-MIL). We propose two specialized attention mechanisms designed to effectively capture the relationships between different types of instances. We innovatively integrate these attention mechanisms with conjunctive pooling applied to the two-dimensional structure at different levels (i.e., bag- and instance-level), enabling TAIL-MIL to effectively pinpoint both the timing and causative multivariate factors of anomalies. We provide theoretical evidence demonstrating TAIL-MIL's efficacy in detecting instances with two-dimensional structures. Furthermore, we empirically validate the superior performance of TAIL-MIL over the state-of-the-art MIL methods and multivariate time-series anomaly detection methods.

Requirements

• torch
• numpy==1.21.6
• pandas==1.3.5
• tqdm==4.64.0
• sklearn==1.0.2

Dataset

5-ECK-2022 (Seasonal)

• In this study, to evaluate the performance of anomaly cause identification, the 5-ECK-2022 dataset is divided into seasonal intervals, and anomalies are labeled using the same labeling method. Additionally, further labeling is performed for the energy sources responsible for the anomalies.

SMD Preprocessing

• The experimental dataset for this study, the Server Machine Dataset (SMD), should be imported from the link below and placed in the path 'datasets/ServerMachineDataset'.
  • link: https://github.com/NetManAIOps/OmniAnomaly/tree/master/ServerMachineDataset
  • Since we only have one example, we can only see the results of one experiment at this time.
• Once you have placed the dataset, you will need to preprocess the SMD using the commands below.

python preprocess.py --dataset SMD

TAIL-MIL (Supervised Version)

• This is the TAIL-MIL model code that performs training with bag-level label data.

TAIL-MIL (Surrogate Model Version)

• This is the Surrogate version of the TAIL-MIL model code, which learns based on reconstruction error.