New CoGenCast framework merges LLMs and flow-matching for time series forecasting

Researchers have introduced CoGenCast, a novel framework designed for time series forecasting that combines autoregressive large language models with a flow-matching mechanism. This hybrid approach aims to capture both the semantic understanding of contextual conditions and the stochastic modeling of continuous temporal dynamics, which are often inadequately handled by existing methods alone. The framework reconfigures pre-trained decoder-only LLMs to enable bidirectional context encoding and causal representation generation, integrating flow-matching for temporal evolution and stochastic dynamics. CoGenCast also supports multimodal forecasting and unified cross-domain training, demonstrating competitive performance in extensive experiments. AI

IMPACT Introduces a novel hybrid approach for time series forecasting by combining LLMs with flow-matching, potentially improving accuracy and multimodal capabilities.

RANK_REASON The cluster contains a research paper detailing a new framework for time series forecasting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• CoGenCast
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv
• ScienceCast
• Yaguo Liu

• paper
• model release

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