Research map / Time series / Foundation models for time series
ORTCL: Towards Continual Learning of Time Series Foundation Models on Streaming Data via Orthogonal Rotation
Li Lin, Xinrui Zhang, Qi Zhang, Shuai Wang, Kaiwen Xia
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 |
|---|---|---|---|
| apply ORTCL orthogonal rotation updates instead of AlphaEdit | Timer · ECL single-domain D1 | MSE: 0.1423 -> 0.1332 | Improved |
| apply ORTCL instead of best baselines | Timer · Single-domain D1, 4 datasets avg | MSE: avg reduction 2.4% vs best baseline | Improved |
| apply ORTCL instead of best baselines | Timer · Single-domain D4, 4 datasets avg | MSE: avg reduction 2.8% vs best baseline | Improved |
| sequentially update with ORTCL instead of full fine-tuning (cross-domain forgetting) | Timer · ECL cross-domain, evaluated after streaming | MSE: FFT 0.1400 -> 0.1628 vs ORTCL 0.1394 -> 0.1454 | Improved |
| sequentially update with ORTCL instead of FFT | Timer · Weather cross-domain | MAE: FFT-fω 0.2961 vs ORTCL-fω 0.2169 | Improved |
| add output-side rotation R (compare w/o R) (ablation) | ORTCL · ECL and Traffic | MAE: w/o R MAE +11.30% ECL, +11.22% Traffic | Improved |
| add input-side rotation S (w/o S ablation) (ablation) | ORTCL · ECL and Traffic | MSE/MAE | Mixed |
| increase truncation rank k from 64 to 1024 (ablation) | ORTCL · ECL and Traffic avg D1-D4 | MSE/MAE | No clear effect |
| ORTCL vs full fine-tuning per segment | Timer · Single-domain D1-D3 | MSE/MAE | Improved |
| use AlphaEdit parameter-modifying editing instead of ORTCL | Timer · ECL/Traffic D1 hidden representations | representation drift | Hurt |
Closest papers by meaning
- Learning Pattern-Specific Experts for Time Series Forecasting Under Patch-level Distribution Shift (NeurIPS 2025 · similarity 0.86)
- ARM: Refining Multivariate Forecasting with Adaptive Temporal-Contextual Learning (ICLR 2024 · similarity 0.86)
- Learning the Context of Errors: Black-Box Online Adaptation of Time Series Foundation Models (NeurIPS 2026 · similarity 0.86)
- OLinear: A Linear Model for Time Series Forecasting in Orthogonally Transformed Domain (NeurIPS 2025 · similarity 0.85)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.85)
- TimeCHEAT: A Channel Harmony Strategy for Irregularly Sampled Multivariate Time Series Analysis (AAAI 2025 · similarity 0.85)
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.
