Research map / Time series / Foundation models for time series
Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistency
Owen Queen, Thomas Hartvigsen, Teddy Koker, Huan He, Theodoros Tsiligkaridis, Marinka Zitnik
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 |
|---|---|---|---|
| use discrete straight-through estimator masks instead of soft attention masking (ablation) | TimeX · ECG | explanation AUPRC: 0.4014 -> 0.4721 | Improved |
| use discrete STE masks instead of soft masks (ablation) | TimeX · ECG | explanation AUR: 0.1564 -> 0.4457 | Improved |
| align explanation latent space with SimCLR loss instead of MBC (ablation) | TimeX · ECG | explanation AUP: 0.5663 -> 0.4895 | Hurt |
| align explanation latent space with SimCLR loss instead of MBC (ablation) | TimeX · ECG | explanation AUPRC: 0.4721 -> 0.4767 | No clear effect |
| use only the model behavior consistency (MBC) objective (ablation) | TimeX · ECG | explanation AUPRC: 0.4721 -> 0.1615 | Hurt |
| use only the label consistency (LC) objective (ablation) | TimeX · ECG | explanation AUPRC: 0.4721 -> 0.3704 | Hurt |
| combine MBC and LC objectives (ablation) | TimeX · ECG | explanation AUPRC: 0.1615 -> 0.4721 | Improved |
| TimeX instead of strongest baseline | TimeX · synthetic datasets | explanation AUP: 3.07% on FreqShapes, 6.3% on SeqComb-UV, 8.43% on SeqComb-MV, 6.24% on LowVar | Improved |
| TimeX instead of strongest baseline | TimeX · ECG | explanation AUPRC: 0.4182 -> 0.4721 | Improved |
| TimeX instead of Integrated Gradients | TimeX · ECG | explanation AUP: 0.5949 -> 0.5663 | Hurt |
| TimeX instead of IG | TimeX · LowVar | explanation AUPRC: 0.8691 -> 0.8673 | No clear effect |
| TimeX instead of Dynamask under occlusion | TimeX · Boiler | classification AUROC under occlusion: 27.8% average, up to 37.4% at the 0.75 threshold | Improved |
Closest papers by meaning
- TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop (NeurIPS 2025 · similarity 0.89)
- Explain Temporal Black-Box Models via Functional Decomposition (ICML 2024 · similarity 0.88)
- ShapeX: Shapelet-Driven Post Hoc Explanations for Time Series Classification Models (NeurIPS 2025 · similarity 0.88)
- TimeXer: Empowering Transformers for Time Series Forecasting with Exogenous Variables (NeurIPS 2024 · similarity 0.86)
- Explaining Time Series via Contrastive and Locally Sparse Perturbations (ICLR 2024 · similarity 0.86)
- TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents (AAAI 2025 · similarity 0.86)
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.
