ADiff4TPP: Asynchronous Diffusion Models for Temporal Point Processes

2025-04-29

AI Summary

This work addresses the limited long-horizon prediction performance in temporal point process (TPP) modeling. We propose the Asynchronous Diffusion Model (ADM), whose core innovation is a time-adaptive asynchronous noise scheduling mechanism: the noise scale dynamically adjusts according to event occurrence times, enabling the generative process to prioritize early events and thereby strengthen conditional dependencies for distant future predictions. Methodologically, ADM employs conditional flow matching to define a unified training objective, supporting flexible and variable-length observation/prediction windows across diverse scenarios, while jointly modeling event types and inter-arrival times in latent space. Evaluated on multiple benchmark datasets, ADM achieves state-of-the-art performance on both event-type and time-interval prediction tasks, with particularly significant gains in long-horizon forecasting accuracy—consistently outperforming all existing baseline methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Temporal Planning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This work introduces a novel approach to modeling temporal point processes using diffusion models with an asynchronous noise schedule. At each step of the diffusion process, the noise schedule injects noise of varying scales into different parts of the data. With a careful design of the noise schedules, earlier events are generated faster than later ones, thus providing stronger conditioning for forecasting the more distant future. We derive an objective to effectively train these models for a general family of noise schedules based on conditional flow matching. Our method models the joint distribution of the latent representations of events in a sequence and achieves state-of-the-art results in predicting both the next inter-event time and event type on benchmark datasets. Additionally, it flexibly accommodates varying lengths of observation and prediction windows in different forecasting settings by adjusting the starting and ending points of the generation process. Finally, our method shows superior performance in long-horizon prediction tasks, outperforming existing baseline methods.

Problem

• Modeling temporal point processes with asynchronous diffusion models
• Improving long-horizon event prediction accuracy
• Flexibly handling varying observation and prediction windows

Innovation

• Asynchronous noise schedule in diffusion models
• Conditional flow matching for training
• Flexible observation and prediction windows