MSS-PAE: Saving Autoencoder-based Outlier Detection from Unexpected Reconstruction

This repo contains source codes for the paper MSS-PAE: Saving Autoencoder-based Outlier Detection from Unexpected Reconstruction.

Software Requirement

• Python 3.7
• numpy 1.19.5
• pandas 1.3.3
• scikit-learn 0.22
• scipy 1.7.1
• torch 1.9.1

Get started

• Prepare the dataset.

  • Place the dataset in ./dataset/original/
    For example:

    ├── original/
        └── Optdigits/
            ├── Optdigits_data.txt
            └── Optdigits_label.txt
    

    The dataset name should be consistent everywhere in this project.

  • Open ./data_pre.py
    Edit "data_dir", "target_dir", and "dataset" according to your setting in the previous step.

  • Run python ./data_pre.py

• Do main experiments.

  • Open ./run_main.sh
    Edit the experimental settings.
    --gpu: Training on which GPU.
    --dataset: Name of the dataset
    --data_dir: Directory of data
    --epochs: Number of training epochs
    --result_dir: Directory to dump results
    --net: Which network to use: AE; PAE
    --alpha: Hyper-paramter alpha for PAE
    --beta: Hyper-paramter beta for PAE
    --inits: How many random initial states for step 2.

  • Run bash ./run_main.sh

Dataset

All datasets used can be found in OutlierNet.

Citation

@article{TAN2025111467,
    title = {MSS-PAE: Saving Autoencoder-based Outlier Detection from Unexpected Reconstruction},
    journal = {Pattern Recognition},
    volume = {163},
    pages = {111467},
    year = {2025},
    issn = {0031-3203},
    doi = {https://doi.org/10.1016/j.patcog.2025.111467},
    url = {https://www.sciencedirect.com/science/article/pii/S003132032500127X}
}