Unified Flow Matching for Long Horizon Event Forecasting

2025-08-06

AI Summary

To address the inefficiency and error accumulation inherent in autoregressive models for long-horizon marked event sequence forecasting, this paper proposes the first unified flow-matching framework for joint non-autoregressive modeling of event times and types. The method innovatively integrates continuous flow matching—used for temporal dynamics—and discrete flow matching—used for mark (type) prediction—within a neural differential equation formalism that captures continuous-time evolution, enabling efficient parallel generation. Evaluated on six real-world datasets, our approach substantially outperforms state-of-the-art autoregressive and diffusion-based baselines, achieving both superior predictive accuracy and speedups of several-fold in generation latency. This work marks the first successful realization of high-quality, high-efficiency joint generation for long-horizon marked event sequences.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Modeling long horizon marked event sequences is a fundamental challenge in many real-world applications, including healthcare, finance, and user behavior modeling. Existing neural temporal point process models are typically autoregressive, predicting the next event one step at a time, which limits their efficiency and leads to error accumulation in long-range forecasting. In this work, we propose a unified flow matching framework for marked temporal point processes that enables non-autoregressive, joint modeling of inter-event times and event types, via continuous and discrete flow matching. By learning continuous-time flows for both components, our method generates coherent long horizon event trajectories without sequential decoding. We evaluate our model on six real-world benchmarks and demonstrate significant improvements over autoregressive and diffusion-based baselines in both accuracy and generation efficiency.

Problem

• Modeling long horizon marked event sequences efficiently
• Reducing error accumulation in long-range event forecasting
• Joint modeling inter-event times and types non-autoregressively

Innovation

• Non-autoregressive joint event modeling
• Continuous and discrete flow matching
• Coherent long horizon trajectory generation