Research map / Time series / Foundation models for time series
Words Towards Explainability: Caption Label-Free Learning via Dual Loop Agentic Time Series Captioning
Difei Hou, Jiaqi Yue, Chunhui Zhao
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 RL optimization with downstream proxy reward | Qwen2.5-VL-7B as Agent · Three-phase flow fault diagnosis | diagnosis accuracy: 0.3976 -> 0.9743 | Improved |
| add RL optimization with downstream proxy reward | Qwen2.5-7B as Agent · Three-phase flow fault diagnosis | diagnosis accuracy: 0.3647 -> 0.8744 | Improved |
| add RL optimization with proxy reward | Qwen2.5-VL-7B as Agent · Caption quality FActScore | FActScore: 0.081 -> 0.913 | Improved |
| add toolset and reflective agent workflow (no fine-tuning) (ablation) | GPT-5.2 · Caption quality ROUGE-L | ROUGE-L: 0.247 -> 0.318 | Improved |
| add toolset and reflective agent workflow (no fine-tuning) (ablation) | GPT-5.2 · Caption quality FActScore | FActScore: 0.738 -> 0.813 | Improved |
| add toolset and reflective agent workflow (no fine-tuning) (ablation) | GPT-4o · Caption quality FActScore | FActScore: 0.711 -> 0.820 | Improved |
| add toolset and agent workflow without fine-tuning (ablation) | Qwen2.5-7B · Caption quality FActScore | FActScore: 0.102 vs 0.450-0.738 for LLM baselines | No clear effect |
| train on forecasting proxy instead of classification proxy | Ours · Forecasting MAE | MAE: 0.42 -> 0.39 | Improved |
| use captions from diagnosis-trained agent for forecasting instead of GPT-4o captions | GPT-4o captions · Forecasting MAE | MAE: 0.61 -> 0.42 | Improved |
| apply method to gas turbine data | Qwen2.5-VL-7B · Gas turbine fault diagnosis | diagnosis accuracy: 0.2314 -> 0.8872 | Improved |
| apply method to medical retinal signal data | Qwen2.5-VL-7B · Disease diagnosis | accuracy: 0.4213 -> 0.8977 | Improved |
| apply method to general time-series reasoning data | Qwen2.5-VL-7B · Classification | accuracy: 0.4601 -> 0.8343 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ITFormer-7B — ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset (ICML 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VDC-Agent: When Video Detailed Captioners Evolve Themselves via Agentic Self-Reflection (ECCV 2026 · similarity 0.85)
- Show, Control and Tell: A Framework for Generating Controllable and Grounded Captions (CVPR 2019 · similarity 0.85)
- Embodied Image Captioning: Self-supervised Learning Agents for Spatially Coherent Image Descriptions (ICCV 2025 · similarity 0.85)
- NarrativeBridge: Enhancing Video Captioning with Causal-Temporal Narrative (ICLR 2025 · similarity 0.84)
- Noise-Aware Image Captioning with Progressively Exploring Mismatched Words (AAAI 2024 · similarity 0.84)
- CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning (ICLR 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.
