Perturbation Learning Based Anomaly Detection

Authors: Jinyu Cai, Jicong Fan

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical studies on benchmark datasets verify the effectiveness and superiority of our method. In this section, we evaluate the proposed method in comparison to several state-of-the-art anomaly detection methods on two image datasets and two tabular datasets.
Researcher Affiliation Academia 1Fuzhou University, Fuzhou, China 2Shenzhen Research Institute of Big Data, Shenzhen, China 3The Chinese University of Hong Kong, Shenzhen, China
Pseudocode No The paper describes the mathematical formulation of the problem and general network architecture, but does not include any pseudocode or formal algorithm blocks.
Open Source Code No The paper's checklist indicates code is included in supplemental material or URL, but the main text does not provide a direct link or explicit statement about the availability of the source code for the described methodology.
Open Datasets Yes CIFAR-10: CIFAR-10 image dataset is composed of 60,000 images in total, where 50,000 samples for training and 10,000 samples for test. Fashion-MNIST: Fashion MNIST contains 10 different categories of grey-scale fashion style objects. The data is split into 60,000 images for training and 10,000 images for test. Thyroid: Thyroid is a hypothyroid disease dataset... We follow the data split settings of [Zong et al., 2018] to preprocess the data... Arrhythmia: Arrhythmia dataset... Here we also follow the data split settings of [Zong et al., 2018] to preprocess the data.
Dataset Splits No The paper specifies training and test samples for datasets like CIFAR-10 and Fashion-MNIST, and refers to external papers for data split settings for others, but it does not explicitly state the proportions or counts for a validation set within its main text.
Hardware Specification Yes Note that we run all experiments on NVIDIA RTX3080 GPU with 32GB RAM, CUDA 11.0 and cu DNN 8.0.
Software Dependencies Yes CUDA 11.0 and cu DNN 8.0.
Experiment Setup Yes For image datasets (CIFAR-10 and Fashion-MNIST), we utilize the Le Net-based CNN to construct the classifier... And we apply the MLP-based VAE to learn the noise for data. For tabular datasets (Thyroid and Arrhythmia), we both use the MLP-based classifier and VAE in practice, and we train them by Adam optimizer with learning rate 0.001. Besides, λ is set to 3 for Thyroid and 2 for Arrhythmia.