Rare-Mark-Aware Next Event Prediction In Marked Event Streams

ICLR 2025 Conference Submission12856 Authors

Keywords: Marked Temporal Point Process

TL;DR: We propose the Rare-Mark-Aware Next Event Prediction problem (RM-NEP) to give rare marks equal opportunity as frequent marks and devise IFNMTPP to efficiently and accurately solve RM-NEP.

Abstract: In marked event streams, Marked Temporal Point Process (MTPP) is central to predicting when and what mark the next event will occur based on the history. In various real-world applications, the mark distribution is significantly imbalanced, i.e., some marks are frequent, and others are rare. We unveil that such imbalance can cause the rare mark missing issue when predicting the next event – frequent marks are dominant, and rare marks often have no chance. However, rare marks can be essential in some applications (e.g., the occurrence of a 7-magnitude earthquake), and missing such rare marks in the next event prediction is risky. To address this issue, we tackle a novel Rare-mark-aware Next Event Prediction problem (RM-NEP), answering two questions for each mark m: “what is the probability that the mark of the next event is m?, and if m, when will the next event happen?”. Solving RM-NEP gives rare marks equal opportunity as frequent marks in the next event prediction. This guarantees that rare marks are always included in the predicted results. Moreover, RM-NEP allows arbitrary number of rare marks samples for time prediction without interference from frequent marks, ensuring the time prediction is accurate. To solve RM-NEP effectively, we first unify the improper integration of two different functions into one and then develop a novel Integral-free Neural Marked Temporal Point Process (IFNMTPP) to approximate the target integral directly. Extensive experiments on real-world and synthetic datasets demonstrate the superior performance of our solution for RM-NEP against various baselines.

Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12856