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..
Channel Matters: Estimating Channel Influence for Multivariate Time Series
Authors: Muyao Wang, Zeke Xie, Bo Chen, Hongwei Liu, James Kwok
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the effectiveness of Ch Inf and Ch Inf-based methods in critical MTS analysis tasks, such as MTS anomaly detection and MTS data pruning. Specifically, our Ch Inf-based methods rank top-1 among all methods for comparison, while previous influence functions do not perform well on MTS anomaly detection tasks and MTS data pruning problem. This fully supports the superiority and necessity of Ch Inf. |
| Researcher Affiliation | Academia | Muyao Wang1,2 EMAIL Zeke Xie3 EMAIL Bo Chen1,2 EMAIL Hongwei Liu1,2 EMAIL James Kwok4 EMAIL 1Intitute of Information Sensing 2School of Electronic Engineering, Xidian University 3Hong Kong University of Science and Technology (Guangzhou) 4Hong Kong University of Science and Technology |
| Pseudocode | Yes | Algorithm 1 Ch Inf based anomaly detection |
| Open Source Code | Yes | Code is available at https://github.com/flare200020/Chinf. |
| Open Datasets | Yes | All the datasets used in our experiments are real-world and open-source MTS datasets. We conduct model comparisons across five widely-used anomaly detection datasets: SMD[40], MSL [41], SMAP [41], SWa T [42], and WADI [19]. In this experiment, we used three benchmark datasets with a large number of channels for testing: Electricity with 321 channels, Solar-Energy with 137 channels, and Traffic with 821 channels. |
| Dataset Splits | Yes | The detailed information of the datasets can be found in Table. 5. [Table 5 contains 'Train' and 'Test' columns with numerical sizes for various datasets]. The detailed information of these datasets can be found in the Table 6. [Table 6 contains 'Datasize' column with comma-separated numbers like '(36601, 5161, 10417)' implying train, validation, test splits]. |
| Hardware Specification | Yes | All experiments were implemented using Py Torch and conducted on a single NVIDIA Ge Force RTX 3090 24GB GPU. |
| Software Dependencies | No | All experiments were implemented using Py Torch and conducted on a single NVIDIA Ge Force RTX 3090 24GB GPU. [The specific version of PyTorch is not mentioned.] |
| Experiment Setup | Yes | For anomaly detection: Models were trained using the SGD optimizer with Mean Squared Error (MSE) loss. For both of them, when trained in reconstructing mode, we used a time window of size 10. For channel pruning: Models were trained using the Adam optimizer with Mean Squared Error (MSE) loss. The input length is 96 and the predicted length is 96. |