Unsupervised Anomaly Detection in The Presence of Missing Values

Authors: Feng Xiao, Jicong Fan

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on datasets with manually constructed missing values and inherent missing values demonstrate that our proposed method effectively mitigates the imputation bias and surpasses the baseline methods significantly.
Researcher Affiliation Academia 1The Chinese University of Hong Kong, Shenzhen, China 2Shenzhen Research Institute of Big Data, Shenzhen, China
Pseudocode No The paper describes the proposed method and its implementation details in Section 3, but it does not include a dedicated pseudocode or algorithm block.
Open Source Code Yes The source code of our method is available at https:// github.com/jicongfan/Im AD-Anomaly-Detection-With-Missing-Data.
Open Datasets Yes We compare Im AD with impute-then-detect methods on 11 publicly available tabular datasets from various fields... The statistics of all datasets are in Table 1 and a detailed description of all datasets is in Appendix J.
Dataset Splits No In all experiments, only incomplete normal data are used in the training stage, but there are both incomplete normal and abnormal data during the inference.
Hardware Specification Yes ALL experiments were conducted on 20 Cores Intel(R) Xeon(R) Gold 6248 CPU with one NVIDIA Tesla V100 GPU, CUDA 12.0.
Software Dependencies Yes ALL experiments were conducted on 20 Cores Intel(R) Xeon(R) Gold 6248 CPU with one NVIDIA Tesla V100 GPU, CUDA 12.0.
Experiment Setup Yes We use MLPs to construct the three modules of Im AD, Adam [Kingma and Ba, 2015] as the optimizer and set coefficient η of entropy regularization term in Sinkhorn distance to 0.1 in all experiments. Other experimental hyper-parameters are provided in Appendix J. Sensitivity analysis of hyper-parameters is provided in Appendix I.