Edit-Based Flow Matching for Temporal Point Processes

2025-10-07

AI Summary

Existing temporal point process (TPP) models predominantly rely on autoregressive generation, suffering from sequential sampling bottlenecks. While non-autoregressive diffusion models alleviate this issue, they still require numerous discrete editing steps. This paper proposes the first edit-flow-based continuous-time non-autoregressive TPP modeling framework: it jointly models event insertion, deletion, and replacement operations within a continuous-time Markov chain, progressively transforming noisy sequences into realistic event sequences. Crucially, it is the first to embed discrete edit operations into the flow matching paradigm, enabling end-to-end optimization by learning instantaneous edit rates. The method drastically reduces generation steps, eliminating autoregressive dependencies. It achieves state-of-the-art performance and enhanced flexibility on both unconditional and conditional TPP generation tasks.

Technology Category

• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Temporal point processes (TPPs) are a fundamental tool for modeling event sequences in continuous time, but most existing approaches rely on autoregressive parameterizations that are limited by their sequential sampling. Recent non-autoregressive, diffusion-style models mitigate these issues by jointly interpolating between noise and data through event insertions and deletions in a discrete Markov chain. In this work, we generalize this perspective and introduce an Edit Flow process for TPPs that transports noise to data via insert, delete, and substitute edit operations. By learning the instantaneous edit rates within a continuous-time Markov chain framework, we attain a flexible and efficient model that effectively reduces the total number of necessary edit operations during generation. Empirical results demonstrate the generative flexibility of our unconditionally trained model in a wide range of unconditional and conditional generation tasks on benchmark TPPs.

Problem

• Generalizing diffusion models for temporal point processes using edit operations
• Learning instantaneous edit rates within continuous-time Markov chain framework
• Reducing necessary edit operations during event sequence generation

Innovation

• Edit Flow process with insert delete substitute operations
• Learning instantaneous edit rates in continuous-time Markov chain
• Reduces total edit operations during generation process