Research map / Time series / Foundation models for time series
TsLLM: Augmenting LLMs for General Time Series Understanding and Prediction
Felix Parker, Nimeesha Chan, Chi Zhang, Kimia Ghobadi
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 |
|---|---|---|---|
| use larger LLM backbone: SmolLM2 135M to Qwen 2.5 7B (ablation) | TsLLM · ECG-QA (ablation) | average accuracy: 58.4 -> 65.0 | Improved |
| use Llama 3.2 3B backbone instead of SmolLM2-360M (ablation) | TsLLM · ECG-QA (ablation) | average accuracy: 59.6 -> 62.5 | Improved |
| use cross attention adapter instead of MLP adapter (ablation) | TsLLM · ECG-QA (ablation) | average accuracy: 59.3 -> 59.6 | No clear effect |
| use cross attention adapter instead of linear adapter (ablation) | TsLLM · ECG-QA (ablation) | average accuracy: 55.8 -> 59.6 | Improved |
| use continuous VAE instead of VQ-VAE encoder (ablation) | TsLLM · ECG-QA (ablation) | average accuracy: 52.4 -> 59.6 | Improved |
| reduce patch size from 32 to 16 (ablation) | TsLLM · ECG-QA encoder ablation | accuracy: 55.9 -> 54.1 | Hurt |
| increase patch size from 32 to 64 (ablation) | TsLLM · ECG-QA encoder ablation | accuracy: 55.9 -> 53.5 | Hurt |
| use plain autoencoder instead of VAE (lower reconstruction MSE) (ablation) | TsLLM · ECG-QA encoder ablation | accuracy: 55.9 -> 54.0 | Hurt |
| use VQ-VAE objective instead of VAE (ablation) | TsLLM · ECG-QA encoder ablation | accuracy: 55.9 -> 51.8 | Hurt |
| increase VAE beta from 1.5 to 1.8 (ablation) | TsLLM · ECG-QA encoder ablation | accuracy: 55.9 -> 55.3 | No clear effect |
| increase encoder layers from 16 to 48 (ablation) | TsLLM · ECG-QA encoder ablation | accuracy: 55.9 -> 54.9 | Hurt |
| use TsLLM instead of Q-Heart | TsLLM · ECG-QA | average accuracy: 61.4 -> 65.0 | Improved |
Closest papers by meaning
- Time-LLM: Time Series Forecasting by Reprogramming Large Language Models (ICLR 2024 · similarity 0.93)
- TimeXL: Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop (NeurIPS 2025 · similarity 0.92)
- AutoTimes: Autoregressive Time Series Forecasters via Large Language Models (NeurIPS 2024 · similarity 0.91)
- STEM-LTS: Integrating Semantic-Temporal Dynamics in LLM-driven Time Series Analysis (AAAI 2025 · similarity 0.91)
- TEST: Text Prototype Aligned Embedding to Activate LLM's Ability for Time Series (ICLR 2024 · similarity 0.91)
- TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents (AAAI 2025 · similarity 0.91)
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.
