In this study, we propose STAR, a novel STate-aware Adapter for Time Series Foundation Models (TSFMs) in the Multivariate Time Serie Anomaly Detection (MTSAD) to better model state variables during the fine-tuning phase. First, to better extract features from state variables, we designed a State Extractor capable of simultaneously capturing both the categorical information and temporal patterns of state variables. This module incorporates an Identity-guided State Encoder which establishes a small-scale State Memory and retrieves a composite memory’s representation to encode categorical information guided by both Variable Identity and State Identity. This method effectively modeling the complex semantics of state variables. Second, we propose a Conditional Bottleneck Adapter. By dynamically generating low-rank adaptation parameters and bottleneck size conditioned on the state variables, this module better integrates the condition-based influence of state variables into the backbone. Building upon the aforementioned modules, we can already perform state-aware fine-tuning. Furthermore, we have developed a Numeral-State Matching module to more effectively detect anomalies contained within the state variables

The below figure provides a overview of STAR's pipeline.

1. Installation:

Given a python environment (note: this project is fully tested under python 3.8), install the dependencies with the following command:

pip install -r requirements.txt

2. Data preparation

Prepare Data. You can obtain the well pre-processed datasets from Google Drive. Then place the downloaded data under the folder ./dataset.

3. Checkpoints preparation

You can download the checkpoints from Google Drive. After obtaining the files, follow the steps below to organize them: For pre-train models (DADA, UniTS, Moment and Timer), move the files to the folder ts_benchmark/baselines/pre_train/checkpoints. Ensure the files are placed in the correct directories for proper functionality.

4. Train and evaluate model

We provide the experiment scripts for all benchmarks under the folder ./scripts/multivariate_detection.

5. Evaluate the TSFMs with standard fine-tuing.

sh ./scripts/evaluation/Backbone.sh

6. Evaluate the TSFMs with State-aware AdapteR (STAR).

sh ./scripts/evaluation/STAR.sh