Practical Adversarial Attacks on Spatiotemporal Traffic Forecasting Models

Authors: Fan LIU, Hao Liu, Wenzhao Jiang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on two real-world datasets show that the proposed two-step framework achieves up to 67.8% performance degradation on various advanced spatiotemporal forecasting models.
Researcher Affiliation Academia Fan LIU AI Thrust&RBM, The Hong Kong University of Science and Technology (Guangzhou) fliu236@connect.hkust-gz.edu.cn&liufan@ust.hk Hao LIU AI Thrust, The Hong Kong University of Science and Technology (Guangzhou) Guangzhou HKUST Fok Ying Tung Research Institute CSE, The Hong Kong University of Science and Technology liuh@ust.hk Wenzhao Jiang AI Thrust, The Hong Kong University of Science and Technology (Guangzhou) wjiang431@connect.hkust-gz.edu.cn
Pseudocode Yes Algorithm 1: Adversarial spatiotemporal attack under the grey-box setting
Open Source Code Yes Our code is available in https://github.com/usail-hkust/Adv-ST.
Open Datasets Yes Datasets. We use two popular real-world datasets to demonstrate the effectiveness of the proposed adversarial attack framework. (1) PEMS-BAY [18] traffic dataset is derived from the California Transportation Agencies (Cal Trans) Performance Measurement System (Pe MS) ranging from January 1, 2017 to May 31, 2017. ... (2) METR-LA [19] is a traffic speed dataset collected from 207 Los Angeles County roadway sensors.
Dataset Splits Yes For evaluation, all datasets are chronologically ordered, we take the first 70% for training, the following 10% for validation, and the rest 20% for testing.
Hardware Specification Yes All experiments are implemented with Py Torch and performed on a Linux server with 4 RTX 3090 GPUs.
Software Dependencies No The paper states "All experiments are implemented with Py Torch" but does not specify the version number of PyTorch or any other software dependencies.
Experiment Setup Yes Implementation details. All experiments are implemented with Py Torch and performed on a Linux server with 4 RTX 3090 GPUs.The traffic speed is normalized to [0, 1]. The input length T and output length τ are set to 12. We select 10% nodes from the whole nodes as the victim nodes, and ε is set to 0.5. The batch size γ is set to 64. The iteration K is set to 5, and the step size α is set to 0.1.