Reproduce “CausalTAD: Causal Implicit Generative Model for Debiased Online Trajectory Anomaly Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenbin Li, Di Yao, Chang Gong, Xiaokai Chu, Quanliang Jing, Xiaolei Zhou, Yuxuan Zhang, Yunxia Fan and 1 more
A causal implicit generative model, CausalTAD, enables debiased online trajectory anomaly detection by mitigating road network preference bias, demonstrating 10.6% to 32.7% higher ROC-AUC and PR-AUC on out-of-distribution data compared to baselines.
Key insights
- Road network preference acts as a hidden confounder, creating spurious correlations between Source-Destination pairs and trajectories that bias existing anomaly detection models.
- Applying Pearl's do-calculus provides a theoretically sound method to derive a debiased anomaly criterion, P(T|do(C)), which can improve generalization to out-of-distribution data.
- The causal effect can be efficiently approximated by factorizing it into a trajectory likelihood and a segment-wise scaling factor, allowing for practical online anomaly detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.