Research map / Time series / Foundation models for time series
ReAugment: Targeted Few-Shot Time Series Augmentation via Model Zoo-Guided Reinforcement Learning
Haochen Yuan, Yutong Wang, Yihong Chen, Yunbo Wang, Xiaokang Yang
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 |
|---|---|---|---|
| augment with ReAugment on iTransformer in few-shot setting | iTransformer · Traffic few-shot | MSE: 0.466 -> 0.427 | Improved |
| add Gaussian, Convolve, TimeGAN or ADA augmentation | iTransformer · ETTh1 few-shot | MSE: 0.411 -> 0.416 (Gaussian), 0.419 (TimeGAN) | Hurt |
| add ReAugment to PatchTST | PatchTST · Traffic few-shot | MSE: 0.541 -> 0.520 | Improved |
| add ReAugment to DLinear | DLinear · Traffic few-shot | MSE: 0.724 -> 0.664 | Improved |
| augment only top 50% high-variance anchors rather than all data (ablation) | ReAugment · few-shot, 8 datasets | FMSE | Improved |
| further filter high-variance anchors by absolute error (ablation) | ReAugment · ETTh1 few-shot | MSE: 0.406 / 0.404 vs 0.402 (all high-var) | No clear effect |
| fine-tune the prior network instead of decoder as RL policy (ablation) | ReAugment · ETTh1 few-shot | MSE: 0.409 -> 0.402 | Improved |
| fine-tune the prior network instead of the full VMAE as RL policy (ablation) | ReAugment · ETTm1 few-shot | MSE: 0.440 -> 0.436 | Improved |
| use GRPO RL finetuning instead of none, REINFORCE or direct backprop (ablation) | ReAugment · ETTh1, ETTm1, Weather, Electricity, Traffic few-shot | MSE | Improved |
| set VMAE mask rate to 0.3 instead of 0 (ablation) | ReAugment · ETTh2 few-shot | MSE: 0.321 -> 0.301 | Improved |
| increase model zoo size K (ablation) | ReAugment · few-shot, 8 datasets | FMSE | Improved |
| train on low-variance half instead of high-variance half of training data (ablation) | iTransformer · ETTm1 | MSE: 0.359 -> 0.329 | Improved |
Closest papers by meaning
- LangTime: A Language-Guided Unified Model for Time Series Forecasting with Proximal Policy Optimization (ICML 2025 · similarity 0.87)
- Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting (NeurIPS 2026 · similarity 0.86)
- Bootstrapped Exploration with Causal Reasoning: A Training Paradigm for Adaptive Forecasting Agent (ICML 2026 · similarity 0.86)
- Predict, Refine, Synthesize: Self-Guiding Diffusion Models for Probabilistic Time Series Forecasting (NeurIPS 2023 · similarity 0.85)
- Forecasting in Offline Reinforcement Learning for Non-stationary Environments (NeurIPS 2025 · similarity 0.85)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.84)
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.
