CoGenCast: A Coupled Autoregressive-Flow Generative Framework for Time Series Forecasting

2026-02-03

AI Summary

This work addresses the challenge that existing methods struggle to jointly model semantic context and continuous stochastic dynamics in time series. To this end, we propose a hybrid generative framework that reconfigures the attention topology of pretrained decoder-only large language models to support an encoder-decoder forecasting architecture, while integrating a flow-matching mechanism to explicitly capture the continuous stochastic evolution process. This approach enables unified cross-domain training and multimodal prediction, achieving substantial performance gains over current baselines across multiple benchmark datasets. The results demonstrate superior forecasting accuracy and strong generalization capabilities, highlighting the effectiveness of our joint modeling of semantic and dynamic components.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Natural Language Processing: Generation
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Time series forecasting can be viewed as a generative problem that requires both semantic understanding over contextual conditions and stochastic modeling of continuous temporal dynamics. Existing approaches typically rely on either autoregressive large language models (LLMs) for semantic context modeling or diffusion-like models for continuous probabilistic generation. However, neither method alone can adequately model both aspects simultaneously. In this work, we propose CoGenCast, a hybrid generative framework that couples pre-trained LLMs with flow-matching mechanism for effective time series forecasting. Specifically, we reconfigure pre-trained decoder-only LLMs into a native forecasting encoder-decoder backbone by modifying only the attention topology, enabling bidirectional context encoding and causal representation generation. Building on this, a flow-matching mechanism is further integrated to model temporal evolution, capturing continuous stochastic dynamics conditioned on the autoregressively generated representation. Notably, CoGenCast naturally supports multimodal forecasting and cross-domain unified training. Extensive experiments on multiple benchmarks show that CoGenCast consistently outperforms previous compared baselines. Code is available at https://github.com/liuyaguo/_CoGenCast.

Problem

• time series forecasting
• semantic context modeling
• stochastic dynamics
• autoregressive models
• generative modeling

Innovation

• autoregressive LLM
• flow-matching
• time series forecasting
• hybrid generative framework
• attention topology