[Paper (TBA)], [Slides]

Dataset

Download NUMOSIM dataset from https://osf.io/sjyfr/ and add the contents to data/NUMOSIM/.

Requirements

Try to use a GPU with more than 24 GB VRAM for training ideally.

Use conda to create a virtual environment and pip to install the requirements:

Environment Setup

conda create --name icad python==3.10.13
conda activate icad
pip install -r requirements.txt

Preprocess

Run python -m utils.preprocess inside ICAD directory.

Training Process

Run python main.py --task next_prediction inside ICAD directory. The backbone of the code is based on TrajGPT implementation.

References

@inproceedings{azarijoo2025icad,
  author    = {Bita Azarijoo and Maria Despoina Siampou and John Krumm and Cyrus Shahabi},
  title     = {ICAD: A Self-Supervised Autoregressive Approach for Multi-Context Anomaly Detection in Human Mobility Data},
  booktitle = {Proceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems},
  year      = {2025},
}