tabular-backdoors

tabular-backdoors

Code repository for Master thesis on backdoor attacks on transformer-based DNNs for tabular data.

Models used

• TabNet, https://arxiv.org/pdf/1908.07442.pdf, (used implementation from https://github.com/dreamquark-ai/tabnet)
• FT-Transformer, https://arxiv.org/pdf/2106.11959.pdf, (used implementation from https://github.com/Yura52/tabular-dl-revisiting-models)
• SAINT, https://arxiv.org/pdf/2106.01342.pdf, (used implementation from https://github.com/somepago/saint)

Data used

• Forest Cover Type (CovType), http://archive.ics.uci.edu/ml/datasets/covertype
• Lending Club Loan (LOAN), https://www.kaggle.com/datasets/wordsforthewise/lending-club and https://www.kaggle.com/datasets/adarshsng/lending-club-loan-data-csv?select=LCDataDictionary.xlsx
• Higgs Boson (HIGGS), https://archive.ics.uci.edu/ml/datasets/HIGGS

Overview

tabular-backdoors           # Project directory
├── data                    # Contains datasets and preprocessing notebooks
├── ExpCleanLabel           # Experiment code for Clean Label Attack
├── ExpInBounds             # Experiment code for In Bounds Trigger
├── ExpTriggerPosition      # Experiment code for Trigger Position based on feature importance
├── ExpTriggerSize          # Experiment code for Trigger Size
├── SAINT                   # SAINT model code
├── FTtransformer           # FT-Transformer model code
└── Notebooks               # Other (smaller or parts of) experiments in the form of notebooks
    ├── FeatureImportances  # Notebooks to calculate feature importance scores and rankings
    └── Defences            # Notebooks on defences against our attacks

Usage

Install and enable environment

virtualenv tabularbackdoor
source tabularbackdoor/bin/activate
pip install -r requirements.txt

# To run the notebooks you also need:
pip install notebook

Download and preprocess data

1. Download accepted_2007_to_2018Q4.csv from https://www.kaggle.com/datasets/wordsforthewise/lending-club and place in data/LOAN/
2. Download LCDataDictionary.xlsx from https://www.kaggle.com/datasets/adarshsng/lending-club-loan-data-csv?select=LCDataDictionary.xlsx and place in data/LOAN/
3. Download HIGGS.csv.gz from https://archive.ics.uci.edu/ml/datasets/HIGGS and extract HIGGS.csv to data/HIGGS
4. Run all four notebooks under data/preprocess to generate the .pkl files containing the datasets for the experiments

Run main experiments

Run the shell script in any of the Exp* folders from the project root with the Python filename (without extension) as argument. Output will be logged to the output folder.

• NOTE: starting an experiment will override the previous log file of the same experiment.
• NOTE: depending on the machine, you might want to edit the GPU used to train each model. To do so, edit the cuda:x string (located somewhere on top) in each .py file.

Example:

bash ExpTriggerSize/run_experiment.sh TabNet_CovType_1F_OOB

To live view the log of a running experiment, use tail -f with the logfile as argument in a new terminal:

tail -f output/triggersize/TabNet_CovType_1F_OOB.log

View results of main experiments

Output logs are found in the output/ folder. All logs end with a section EASY COPY PASTE RESULTS: where you can copy the resulting lists containing the ASR and BA for each run.

Run notebooks (Defences and FeatureImportance calculations)

See the Notebooks/ folder for other (smaller or parts of) experiments in the form of notebooks. To run the defences, you must first run the appropiate CreateModel Notebook to create a backdoored model and dataset which can then be analyzed with the other Notebooks. For Fine-Pruning defence, there is a dedicated subfolder in the Notebooks/Defences folder with notebooks to train, prune and finetune FTT.