Variational Counterfactual Intervention Planning to Achieve Target Outcomes
Authors: Xin Wang, Shengfei Lyu, Chi Luo, Xiren Zhou, Huanhuan Chen
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both synthetic and real-world datasets show that VCIP significantly outperforms existing methods in target achievement accuracy. In this section, we conduct comprehensive experiments to evaluate the performance of VCIP against several state-of-the-art baselines. |
| Researcher Affiliation | Academia | 1University of Science and Technology of China 2Nanyang Technological University. Correspondence to: Huanhuan Chen <EMAIL>, Shengfei Lyu <EMAIL>. |
| Pseudocode | Yes | Algorithm 1 Optimize Intervention Sequence |
| Open Source Code | Yes | Code available at: https://github.com/ wangxin0126/VCIP-ICML |
| Open Datasets | Yes | The tumor dataset simulates lung cancer treatment dynamics through a pharmacokineticpharmacodynamic framework (Geng et al., 2017). This dataset serves as a benchmark for evaluating causal inference in sequential decision-making (Lim et al., 2018; Bica et al., 2020). For real-world evaluation, we leverage the Medical Information Mart for Intensive Care III (MIMIC-III) database (Johnson et al., 2016), which contains de-identified electronic health records from intensive care unit patients. |
| Dataset Splits | Yes | The dataset comprises 1,000 training trajectories, 100 validation sequences, and 100 test cases across different confounding levels γ. This dataset undergoes a strategic partition into training (70%), validation (15%), and testing (15%) subsets. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware used to run its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. |
| Experiment Setup | Yes | The hyperparameter settings used in our experiments are detailed in Table 7. For all baseline models including RMSN, CRN, CT, and ACTIN, we followed the hyperparameter optimization strategy and ranges consistent with those reported in (Wang et al., 2024). Specifically, hyperparameter optimization was conducted through random grid search. Readers can refer to (Wang et al., 2024) for detailed hyperparameter configurations of these baseline models. |