Counterfactual History Distillation on Continuous-time Event Sequences

Sishun Liu, KE DENG, Yan Wang, Xiuzhen Zhang

Keywords: Counterfactual Analysis, Marked Temporal Point Process

TL;DR: We devise a new approach to search for essential information in history from the perspective of the trained MTPP model by counterfactual analysis.

Abstract: This study aims to distill history events that have essential information for predicting subsequent events with counterfactual analysis. The problem is named Counterfactual History Distillation (CHD). CHD distills a minimum set of events from history, based on which the distribution provided by a trained MTPP model fits the events observed later, and the distribution based on the remaining events in history cannot. It can help understand what event marks may have more influence on the occurrence of future events and what events in history may have a causal relationship with the events observed later. This study proposes a robust solution for CHD, called MTPP-based Counterfactual History Distiller (MTPP-CHD). MTPP-CHD learns to select the optimal event combination from history for the events observed later. Experiment results demonstrate the superiority of MTPP-CHD by outperforming baselines in terms of distillation quality and processing speed.

Primary Area: causal reasoning

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: 12894