Academic

Bridging Discrete Marks and Continuous Dynamics: Dual-Path Cross-Interaction for Marked Temporal Point Processes

arXiv:2603.11462v1 Announce Type: new Abstract: Predicting irregularly spaced event sequences with discrete marks poses significant challenges due to the complex, asynchronous dependencies embedded within continuous-time data streams.Existing sequential approaches capture dependencies among event tokens but ignore the continuous evolution between events, while Neural Ordinary Differential Equation (Neural ODE) methods model smooth dynamics yet fail to account for how event types influence future timing.To overcome these limitations, we propose NEXTPP, a dual-channel framework that unifies discrete and continuous representations via Event-granular Neural Evolution with Cross-Interaction for Marked Temporal Point Processes. Specifically, NEXTPP encodes discrete event marks via a self-attention mechanism, simultaneously evolving a latent continuous-time state using a Neural ODE. These parallel streams are then fused through a crossattention module to enable explicit bidirectional interac

Y
Yuxiang Liu, Qiao Liu, Tong Luo, Yanglei Gan, Peng He, Yao LIu · · 1 min read · 4 views
#cs.LG #cs.AI

arXiv:2603.11462v1 Announce Type: new Abstract: Predicting irregularly spaced event sequences with discrete marks poses significant challenges due to the complex, asynchronous dependencies embedded within continuous-time data streams.Existing sequential approaches capture dependencies among event tokens but ignore the continuous evolution between events, while Neural Ordinary Differential Equation (Neural ODE) methods model smooth dynamics yet fail to account for how event types influence future timing.To overcome these limitations, we propose NEXTPP, a dual-channel framework that unifies discrete and continuous representations via Event-granular Neural Evolution with Cross-Interaction for Marked Temporal Point Processes. Specifically, NEXTPP encodes discrete event marks via a self-attention mechanism, simultaneously evolving a latent continuous-time state using a Neural ODE. These parallel streams are then fused through a crossattention module to enable explicit bidirectional interaction between continuous and discrete representations. The fused representations drive the conditional intensity function of the neural Hawkes process, while an iterative thinning sampler is employed to generate future events. Extensive evaluations on five real-world datasets demonstrate that NEXTPP consistently outperforms state-of-the-art models. The source code can be found at https://github.com/AONE-NLP/NEXTPP.

Executive Summary

This article proposes a novel framework, NEXTPP, to predict irregularly spaced event sequences with discrete marks, by unifying discrete and continuous representations. It leverages a self-attention mechanism for event marks and a Neural ODE for continuous-time state evolution, with a cross-attention module to enable interaction between the two. The framework is evaluated on five real-world datasets, consistently outperforming state-of-the-art models. The proposed method addresses the limitations of existing sequential approaches and Neural ODE methods, which either ignore continuous evolution or event-type influences. The framework's performance and adaptability suggest its potential in various applications, such as event prediction and sequence modeling.

Key Points

  • ▸ NEXTPP is a dual-channel framework for marked temporal point processes, unifying discrete and continuous representations.
  • ▸ The framework employs a self-attention mechanism for event marks and a Neural ODE for continuous-time state evolution.
  • ▸ A cross-attention module enables explicit bidirectional interaction between continuous and discrete representations.

Merits

Strength in Predictive Performance

NEXTPP consistently outperforms state-of-the-art models on five real-world datasets, demonstrating its superiority in predicting irregularly spaced event sequences.

Adaptability and Flexibility

The framework's modular design allows for easy adaptation to various applications, such as event prediction and sequence modeling, making it a versatile tool for researchers and practitioners.

Demerits

Computational Complexity

The framework's dual-channel design, involving self-attention and cross-attention mechanisms, may increase computational complexity and require significant computational resources for large-scale applications.

Limited Interpretability

The complex interactions between continuous and discrete representations may limit interpretability of the results, making it challenging to understand the underlying mechanisms driving the predictions.

Expert Commentary

This article presents a significant contribution to the field of marked temporal point processes, addressing the limitations of existing sequential approaches and Neural ODE methods. The proposed NEXTPP framework demonstrates its superiority in predicting irregularly spaced event sequences, making it a valuable tool for researchers and practitioners. However, the framework's computational complexity and limited interpretability are potential concerns. Further research is needed to address these issues and explore the framework's potential in various applications. The article's contributions and implications suggest that NEXTPP has the potential to become a prominent method in the field of event sequence modeling and prediction.

Recommendations

  • ✓ Recommendation 1: Researchers should investigate the framework's performance on larger and more diverse datasets to assess its scalability and generalizability.
  • ✓ Recommendation 2: Practitioners should consider the framework's computational complexity and potential limitations in interpretability when applying it to real-world applications.

Sources

Related Articles