Research map / Time series / Foundation models for time series
BEDTime: A Unified Benchmark for Automatically Describing Time Series
Medhasweta Sen, Zachary Gottesman, Jiaxing Qiu, C. Bayan Bruss, Nam H Nguyen, Thomas Hartvigsen
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 |
|---|---|---|---|
| give the model a rendered plot instead of text (VLM vs LLM) | GPT-5.4-mini · TRUCE-Stock Recognition | accuracy: .68 -> .79 | Improved |
| give the model a rendered plot instead of text (VLM vs LLM) | GPT-5.4-mini · SUSHI Recognition | accuracy: .89 -> .97 | Improved |
| use LLMTime prompting | GPT-4o · SUSHI Recognition | accuracy: .74 -> .84 | Improved |
| use LLMTime prompting | GPT-4o · NICU-HR Recognition | accuracy: .57 -> .57 | No clear effect |
| double LLM size from 7B to 14B | Qwen2.5 · SUSHI Recognition | accuracy: .51 -> .51 | No clear effect |
| add chain-of-thought prompting (ablation) | LLMs · Recognition and Differentiation | mean accuracy: +0.03 | Improved |
| add chain-of-thought prompting (ablation) | Qwen2.5-7B · Recognition (average) | accuracy: +11.85% | Improved |
| add chain-of-thought prompting (ablation) | GPT-4o · TRUCE-Stock | positional bias: 18.1% (R) and 11.7% (D) reduction | Improved |
| scale interpolated values by 100 to make integers (ablation) | LLMs except Qwen2.5-14B · interpolated long series | accuracy | Improved |
| mask up to 25% of values (ablation) | LLMs · Differentiation | accuracy | No clear effect |
| mask 50% or more of values (ablation) | LLMs, TSLMs · Differentiation | accuracy | Hurt |
| add Gaussian noise (ablation) | text-only LLMs · Differentiation | accuracy | Hurt |
Closest papers by meaning
- TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop (NeurIPS 2025 · similarity 0.89)
- ChatTime: A Unified Multimodal Time Series Foundation Model Bridging Numerical and Textual Data (AAAI 2025 · similarity 0.89)
- AutoTimes: Autoregressive Time Series Forecasters via Large Language Models (NeurIPS 2024 · similarity 0.88)
- LAST SToP for Modeling Asynchronous Time Series (ICML 2025 · similarity 0.88)
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.88)
- ITFormer: Bridging Time Series and Natural Language for Multi-Modal QA with Large-Scale Multitask Dataset (ICML 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.
