Treatment Rule Optimization Under Counterfactual Temporal Point Processes with Latent States
Keywords: counterfactual reasoning, temporal point processes, latent confounder, rule learning
Abstract: In high-stakes areas like healthcare, retrospective counterfactual analysis—such as evaluating what might have happened if treatments were administered earlier, later, or differently—is vital for refining treatment strategies. This paper proposes a counterfactual treatment optimization framework using temporal point processes to model outcome event sequences. By sampling potential outcome events under new treatment decision rules, our approach seeks to optimize treatment strategies in a counterfactual setting. To achieve accurate counterfactual evaluation of new decision rules, we explicitly introduce latent states into the modeling of temporal point processes. Our method first infers the latent states and associated noise, followed by counterfactual sampling of outcome events. This approach rigorously addresses the complexities introduced by latent states, effectively removing biases in the evaluation of treatment strategies. By proving the identifiability of model parameters in the presence of these states, we provide theoretical guarantees that enhance the reliability and robustness of the counterfactual analysis. By incorporating latent states and proving identifiability, our framework not only improves the accuracy and robustness of treatment decision rules but also offers actionable insights for optimizing healthcare interventions. This method holds significant potential for improving treatment strategies, particularly in healthcare scenarios where patient symptoms are complex and high-dimensional.
Primary Area: interpretability and explainable AI
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: 13574
Loading