Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Multivariate Time Series Anomaly Detection with Idempotent Reconstruction
Authors: Xin Sun, Heng Zhou, Chao Li
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Regarding the latest findings of AD metrics, we evaluated IGAD on various methods with four realworld datasets, and they achieve visible improvements in VUS-PR than their predecessors, demonstrating the effective potential of IGAD for further improvements in MTS AD tasks. Our instructions on integrating IGAD into customized models and example codes are available at https://github.com/Pro Echo1/ Idempotent-Generation-for-Anomaly-Detection-IGAD. The major contributions of our work can be three-folded: (3) Based on our experimental results in VUS-PR, noise-affected verification, and the distributions of anomaly scores, we demonstrate the effectiveness of IGAD in further improving the performance of different models in multivariate time series anomaly detection tasks. |
| Researcher Affiliation | Academia | Xin Sun Zhejiang University EMAIL Heng Zhou Zhejiang University EMAIL Chao Li Zhejiang University EMAIL |
| Pseudocode | Yes | Algorithm 1 Nyquist Criterion Compliance Verification. Listing 1: Python implementation for IGAD. |
| Open Source Code | Yes | Our instructions on integrating IGAD into customized models and example codes are available at https://github.com/Pro Echo1/ Idempotent-Generation-for-Anomaly-Detection-IGAD. We have released the codes and datasets in the public repository https:// github.com/Pro Echo1/Idempotent-Generation-for-Anomaly-Detection-IGAD with included instructions. |
| Open Datasets | Yes | In our experiments, we selected four redesigned public datasets commonly used for MTS AD in [32], including SMD from [50], MSL from [24], PSM from [1] and SMAP from [24]. We have released the codes and datasets in the public repository https:// github.com/Pro Echo1/Idempotent-Generation-for-Anomaly-Detection-IGAD with included instructions. |
| Dataset Splits | No | The latest study on datasets and benchmarks [32] has designed well-organized datasets and searched for optimal hyperparameters in optimizer, learning rate, and weights of existing loss functions for the majority models included in this study. For some of our selected base models, which are not temporarily imported, we set their hyperparameters in their original papers or repositories as optimal ones. Then these models are also integrated into this proposed pipeline to run in a universal data flow. |
| Hardware Specification | Yes | We program our codes with Python 3.8.13, Py Torch 1.13.0, CUDA 11.7 and Ubuntu 18.04 on a single NVIDIA RTX 3090 24GB GPU. All experiments are conducted under the same environments. All experiments are performed with a single NVIDIA RTX 3090 24GB GPU, and 0.52 GB is also acceptable for most hardware conditions for deep learning currently. |
| Software Dependencies | Yes | We program our codes with Python 3.8.13, Py Torch 1.13.0, CUDA 11.7 and Ubuntu 18.04 on a single NVIDIA RTX 3090 24GB GPU. All experiments are conducted under the same environments. |
| Experiment Setup | Yes | Under these settings, λrec and λaux are fixed, then intervals [0.1, 1.0] with a step size of 0.1 for λidem and λtight, as well as [1.1, 1.5] with a step size of 0.1 for α are used for a detailed grid search. Optuna [2] is selected for the search process. Then, we summarize the optimal hyperparameters for each model and each dataset in Tab.5. |