Research map / Time series / Foundation models for time series
Time-PEFT: Temporal and Multichannel Complexity-Based Fine-Tuning for Time-Series Foundation Models
Jihye Na, Patara Trirat, Chanyoung Park, Jae-Gil Lee
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 channel adapter to LoRA (ablation) | LoRA on MOMENT-base · Lorenz | MSE: 0.594 -> 0.417 | Improved |
| add frequency adapter passing only filtered embedding to LoRA (ablation) | LoRA on MOMENT-base · Lorenz | MSE: 0.594 -> 0.544 | Improved |
| keep backbone embedding alongside filtered embedding in frequency adapter (ablation) | LoRA + frequency adapter · Lorenz | MSE: 0.544 -> 0.520 | Improved |
| add frequency adapter on top of LoRA + channel adapter (ablation) | Time-PEFT · Lorenz | MSE: 0.417 -> 0.385 | Improved |
| add Time-PEFT adapters to LoRA | LoRA on MOMENT-base · Complex datasets (7) | relative MSE improvement over LoRA: 5.6 to 38.0 % | Improved |
| add Time-PEFT adapters to LoRA | LoRA on MOMENT-base · Standard datasets | relative MSE improvement over LoRA: -7.6 to 7.1 % | No clear effect |
| use ChannelMixing instead of LoRA | MOMENT-base · Complex datasets | relative MSE improvement over LoRA: -23.6 to 17.0 % | Mixed |
| use FourierFT instead of LoRA | MOMENT-base · Complex datasets | relative MSE improvement over LoRA: -2.1 to -0.1 % | No clear effect |
| vary adapter rank r from 48 to 384 (ablation) | Time-PEFT · MOMENT-base complex datasets | MSE | No clear effect |
| vary top-k from 1 to 5 (ablation) | Time-PEFT · MOMENT-base complex datasets | MSE | No clear effect |
| use rank 48 (h1/16) instead of 384 (ablation) | Time-PEFT · Lorenz MOMENT-base | trainable parameters (M): 4.429 -> 3.138 | Improved |
| Time-PEFT versus full fine-tuning | Full FT · UniTS Lorenz | MSE: 0.161 -> 0.177 | Mixed |
Closest papers by meaning
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.89)
- In-Context Fine-Tuning for Time-Series Foundation Models (ICML 2025 · similarity 0.89)
- TimeBase: The Power of Minimalism in Efficient Long-term Time Series Forecasting (ICML 2025 · similarity 0.88)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.87)
- SEMPO: Lightweight Foundation Models for Time Series Forecasting (NeurIPS 2025 · similarity 0.87)
- TimeCHEAT: A Channel Harmony Strategy for Irregularly Sampled Multivariate Time Series Analysis (AAAI 2025 · similarity 0.87)
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.
