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..
Self-Perturbed Anomaly-Aware Graph Dynamics for Multivariate Time-Series Anomaly Detection
Authors: Jinyu Cai, Yuan Xie, Glynnis Lim, Yifang Yin, Roger Zimmermann, See-Kiong Ng
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across multiple benchmark datasets demonstrate the effectiveness of SPAGD compared to state-of-the-art baselines. |
| Researcher Affiliation | Academia | Jinyu Cai Institute of Data Science National University of Singapore EMAIL Yuan Xie School of Computing National University of Singapore EMAIL Glynnis Lim Institute of Data Science National University of Singapore EMAIL Yifang Yin Institute for Infocomm Research A*STAR, Singapore EMAIL Roger Zimmermann School of Computing National University of Singapore EMAIL See-Kiong Ng Institute of Data Science National University of Singapore EMAIL |
| Pseudocode | Yes | We also provide a detailed algorithm description in the Appendix C. |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The paper has provided open access to the data and code. |
| Open Datasets | Yes | In this paper, we evaluate the proposed SPAGD method on three public time-series datasets collected from different real-world scenarios, including (1) Secure Water Treatment (SWaT) [Mathur and Tippenhauer, 2016], (2) Soil Moisture Active Passive (SMAP) [Hundman et al., 2018], and (3) Mars Science Laboratory (MSL) [Hundman et al., 2018]. |
| Dataset Splits | Yes | For a fair evaluation, all baseline methods were trained and tested using identical data splits for each dataset. ... Table 1 summarizes the main attributes of these datasets, and we also detail the information of each dataset in the Appendix A. |
| Hardware Specification | No | The main body of the paper does not specify the exact hardware (GPU/CPU models, memory) used for running the experiments. While the NeurIPS Paper Checklist indicates that such information is provided, it is not present in the excerpt provided for analysis. |
| Software Dependencies | No | The paper mentions using a Transformer-based model and GAT layers, and describes architectural details (e.g., '8 attention heads and 3 encoder layers', '2 graph attention network (GAT) layers with a latent dimension of 256'). However, it does not provide specific version numbers for any software libraries, frameworks, or programming languages used (e.g., Python, PyTorch, TensorFlow, CUDA versions). |
| Experiment Setup | Yes | For SPAGD, we employed a Transformer-based reconstruction model comprising 8 attention heads and 3 encoder layers for self-perturbation learning. During the spatio-temporal modeling stage, each time series sample first undergoes spatial graph convolution that contains 2 graph attention network (GAT) [Veliหckovi c et al., 2018] layers with a latent dimension of 256. The learned spatial features were subsequently partitioned into 5 equal-length segments for temporal convolution processing. Aggregated spatio-temporal features were then passed through a predictor consisting of two fully-connected layers to produce final anomaly scores. For other experimental settings and training details, please refer to the Appendix B. |