Research map / Time series / Foundation models for time series
TimeGuard: Channel-wise Pool Training for Backdoor Defense in Time Series Forecasting
Quang Duc Nguyen, Siyuan Liang, Yiming Li, Fushuo Huo, Dacheng Tao
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 |
|---|---|---|---|
| adopt channel-wise reliable pool training instead of sample-level filtering (ablation) | TimeGuard · PEMS03, Random attack | FDER: 0.478 -> 0.868 | Improved |
| add Distance-Regularized Loss Selection instead of loss-only selection (ablation) | TimeGuard · PEMS03, Random attack | FDER: 0.607 -> 0.868 | Improved |
| add Neighborhood Diversity Filtering to Stage I pool initialization (ablation) | TimeGuard · PEMS03, Random attack | FDER: 0.853 -> 0.868 | Improved |
| add Reverse-Consistency Filtering to Stage I pool initialization (ablation) | TimeGuard · PEMS03, BackTime attack | FDER: 0.796 -> 0.808 | Improved |
| add NDF and RCF together in Stage I (ablation) | TimeGuard · PEMS03, Random attack | MAEP: 91.780 -> 104.677 | Improved |
| use TimeGuard instead of the strongest baseline PDB | PEMS03 averaged over attacks · PEMS03 | MAEP: 1.96x relative improvement in MAEP | Improved |
| apply sample-level filtering defenses (Spectral, TED, TED++) to TSF | Spectral, TED, TED++ · PEMS03 | FDER: FDER about 0.54 | No clear effect |
| rely on training-loss-only defenses (ABL, ESTI) | ABL, ESTI · PEMS03 | FDER: average FDER 0.497 | No clear effect |
| use inference-time detection (STRIP, TeCo, IBD-PSC) adapted to TSF | TeCo (best) · PEMS03, Weather, ETTm1 | AUROC: 0.551 AUROC and 0.559 F1 for TeCo | No clear effect |
| set neighbourhood size K from small to large (ablation) | TimeGuard · PEMS03, BackTime | FDER: FDER 0.805-0.809 | No clear effect |
| increase scaling factor pi beyond 1.5 (ablation) | TimeGuard · PEMS03, BackTime | FDER | Hurt |
| set pool size beta too small or too large (best 0.5-0.7) (ablation) | TimeGuard · PEMS03, BackTime | FDER: MAEP 39.28 at beta 0.7 vs 33.59 at beta 0.8 (alpha 0.25) | Hurt |
Closest papers by meaning
- BackTime: Backdoor Attacks on Multivariate Time Series Forecasting (NeurIPS 2024 · similarity 0.90)
- LT-Defense: Searching-free Backdoor Defense via Exploiting the Long-tailed Effect (NeurIPS 2024 · similarity 0.84)
- TimePerceiver: An Encoder-Decoder Framework for Generalized Time-Series Forecasting (NeurIPS 2025 · similarity 0.84)
- Backdoor Attacks in Token Selection of Attention Mechanism (ICML 2025 · similarity 0.84)
- PolicyGuard: Towards Test-time and Step-level Adversary Defense for Reinforcement Learning Agent (ICML 2026 · similarity 0.83)
- TimeClaw: A Time-Series AI Agent with Exploratory Execution Learning (NeurIPS 2026 · similarity 0.83)
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.
