Research map / Time series / Long-term forecasting
Beyond Immediate Activation: Temporally Decoupled Backdoor Attacks on Time Series Forecasting
Zhixin Liu, Xuanlin Liu, Sihan Xu, Yaqiong Qiao, Ying Zhang, Xiangrui Cai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add position guidance matrix (Gaussian prior on target position) to the trigger generator (ablation) | TDBA-Inv · ablation, aggregate over datasets | attacked-position MAE M_p^a: 35.34 -> 18.25 | Improved |
| add position guidance matrix to the trigger generator (ablation) | TDBA-Gcn · ablation, aggregate over datasets | attacked-position MAE M_p^a: 30.64 -> 16.94 | Improved |
| add position-aware optimization objective instead of plain MAE regression (ablation) | TDBA-Inv · ablation, aggregate over datasets | attacked-position MAE M_p^a: 29.48 -> 18.25 | Improved |
| add position-aware optimization objective instead of plain MAE regression (ablation) | TDBA-Gcn · ablation, aggregate over datasets | unaffected-position MAE M_p^c: 21.44 -> 17.40 | Improved |
| use TDBA over BackTime for delayed asynchronous target activation | TDBA · Weather (avg) | attacked-position MAE M_p^a: 23.22 -> 3.48 | Improved |
| use TDBA over BackTime on PEMS08 | TDBA-Gcn · PEMS08 | attacked-position MAE M_p^a: 25.10 -> 17.52 | Improved |
| use TDBA-Gcn instead of BackTime on ETTh1 | TDBA-Gcn · ETTh1 | attacked-position MAE M_p^a: 2.74 -> 2.90 | Hurt |
| use TDBA under different target pattern shapes | TDBA-GCN · PEMS03 Autoformer upward trend | attacked-position MAE M_p^a: 20.04 -> 17.70 | Improved |
| stealthiness check with USAD anomaly detection | TDBA · PEMS03, PEMS04, PEMS08 | AUC-ROC: BackTime 0.5747 -> TDBA-Inv 0.4949 on PEMS04 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BackTime — BackTime: Backdoor Attacks on Multivariate Time Series Forecasting (NeurIPS 2024) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- BackTime: Backdoor Attacks on Multivariate Time Series Forecasting (NeurIPS 2024 · similarity 0.92)
- TimeGuard: Channel-wise Pool Training for Backdoor Defense in Time Series Forecasting (ICML 2026 · similarity 0.89)
- Fortifying Time Series: DTW-Certified Robust Anomaly Detection (NeurIPS 2025 · similarity 0.84)
- Timer-XL: Long-Context Transformers for Unified Time Series Forecasting (ICLR 2025 · similarity 0.83)
- Rethinking Channel Dependence for Multivariate Time Series Forecasting: Learning from Leading Indicators (ICLR 2024 · similarity 0.82)
- TimeClaw: A Time-Series AI Agent with Exploratory Execution Learning (NeurIPS 2026 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
