New AI Model Bidirectionally Couples LLMs and Time Series Models

Researchers have introduced TimeOmni-o1, a novel architecture that bidirectionally couples a Time Series Foundation Model (TSFM) with a Large Language Model (LLM). This integration allows for multimodal reasoning and forecasting by enabling the TSFM to encode time series data for the LLM, which then reasons over both text and the compressed series representation. The LLM's reasoning is also fed back into the TSFM to condition forecasts. To support this, a new dataset called MMTR was created, featuring over 500,000 samples across 12 domains for multimodal time series reasoning and forecasting. AI

IMPACT This model could enhance AI capabilities in domains requiring reasoning over both structured time-series data and unstructured text, improving forecasting and analytical tasks.

RANK_REASON The cluster describes a new research paper detailing a novel AI model architecture and dataset. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Hugging Face
• LLMs
• TimeOmni-o1
• Time series foundation models

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →