Temporal point processes (TPPs) are a powerful framework for modeling event sequences with irregular timestamps, such as those commonly found in electronic health records (EHR), which often involve high-dimensional and diverse event types. However, building generative models for such complex datasets comes with several challenges, including addressing sample inefficiency, accurately capturing intricate event patterns, and producing outputs that are both trustworthy and interpretable. In this paper, we present a neuro-symbolic generative model for TPPs based on the Variational Autoencoder (VAE) framework. Our model incorporates a neural-symbolic reasoning layer into the latent space, allowing it to integrate interpretable, logic-based constraints and perform logical reasoning over learned representations. This integration enhances the interpretability of the latent space by embedding logic rules directly into the generative process, enabling structured reasoning and improved decision-making based on underlying data patterns. We validate our model on an ICU EHR dataset, demonstrating its effectiveness in capturing complex event dynamics with irregular timestamps. In addition to improving sample efficiency and accuracy, our model supports the secure and interpretable generation of synthetic event data, making it a valuable tool for healthcare applications where reliability and trustworthiness are critical.
Keywords: temporal point process, neuro-symbolic, generative model
Abstract:
Primary Area: generative models
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: 13231
Loading