Task-Driven Causal Feature Distillation: Towards Trustworthy Risk Prediction
Authors: Zhixuan Chu, Mengxuan Hu, Qing Cui, Longfei Li, Sheng Li
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we conduct experiments on synthetic and real datasets, including causal effect estimation benchmarks and synthetic and real datasets for risk prediction tasks to evaluate the following aspects: (1) Our proposed method based on relational graph construction and adaptive group Lasso PS estimation can ensure the accuracy of causal feature attribution estimation; (2) The precision and recall of the risk prediction task are significantly improved. |
| Researcher Affiliation | Collaboration | Zhixuan Chu1, Mengxuan Hu2, Qing Cui1, Longfei Li1, Sheng Li2 1Ant Group 2University of Virginia |
| Pseudocode | No | No pseudocode or algorithm blocks (labeled as 'Pseudocode', 'Algorithm', or similar) are present in the paper. |
| Open Source Code | No | The paper does not provide any statement or link regarding the public availability of its source code. |
| Open Datasets | Yes | We conduct the causal effect estimation experiments on the News dataset with different interventions and compare our TDCFD model with eleven baselines. Results. We adopt the commonly used evaluation metric, i.e., the error in average treatment effect (ATE) estimation defined as ϵATE = |ATE d ATE|, where d ATE is an estimated ATE. [...] We list the available results reported by the original authors (Schwab, Linhardt, and Karlen 2018). |
| Dataset Splits | No | The paper does not explicitly provide details on train/validation/test dataset splits (e.g., percentages, sample counts, or specific splitting methodologies). While it mentions 'test sets' in table captions, it doesn't define the full data partitioning. |
| Hardware Specification | No | The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments. |
| Software Dependencies | No | The paper mentions various models and techniques (e.g., Logistic Regression, SVM, KNN, RF, XGBoost, DNN, Transformer) but does not provide specific version numbers for any software dependencies or libraries used in the implementation. |
| Experiment Setup | No | The paper does not contain specific experimental setup details such as hyperparameter values (e.g., learning rate, batch size, number of epochs) or specific training configurations. |