Accelerating Shapley Explanation via Contributive Cooperator Selection
Authors: Guanchu Wang, Yu-Neng Chuang, Mengnan Du, Fan Yang, Quan Zhou, Pushkar Tripathi, Xuanting Cai, Xia Hu
ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To demonstrate the effectiveness, we comprehensively evaluate SHEAR across multiple metrics including the absolute error from the ground-truth Shapley value, the faithfulness of the explanations, and running speed. The experimental results indicate SHEAR consistently outperforms state-of-the-art baseline methods across different evaluation metrics |
| Researcher Affiliation | Collaboration | 1Department of Computer Science, Rice University 2Department of Computer Science and Engineering, Texas A&M University 3Meta Platforms, Inc. |
| Pseudocode | Yes | The configuration and pseudo code of SHEAR are given in Figure 1 and Algorithm 1, respectively. Algorithm 1 SHapley Explanation Accele Ration (SHEAR) |
| Open Source Code | Yes | The source code is available at https: //github.com/guanchuwang/SHEAR. |
| Open Datasets | Yes | Dataset: The experiments involve Census Income, German Credit and Cretio datasets from the areas of social media, finance and recommender systems, respectively. More details about the datasets are provided in Appendix F. |
| Dataset Splits | Yes | Table 2: Dataset Statistics Dataset Continuous Categorical Training Validation Testing Census Income 5 8 20838 5210 6513 German Credit 7 9 28934 7234 9043 Cretio 13 26 80000 10000 10000 |
| Hardware Specification | Yes | Table 4: Computing infrastructure for the experiments. Device Attribute Value Computing infrastructure CPU CPU model Apple M1 CPU number 1 Core number 8 Memory size 16GB |
| Software Dependencies | No | The paper mentions using 'Adam optimizer' and 'Deep FM' but does not specify their version numbers or the versions of underlying software libraries like Python, PyTorch, or TensorFlow. |
| Experiment Setup | Yes | Model Training: We adopt 3-layer MLP (multi-layer perceptron) as the classification model for the Census Income and German Credit datasets. To train the model, we adopt Adam optimizer with 10^-3 learning rate... For the Cretio dataset, we use Deep FM (Guo et al., 2017) as the model and adopt Adam optimizer with 10^-4 learning rate... Batch Size 256... Hidden dim. 64 (for MLP) / 32 (for Deep FM)... the reference value of continuous features takes the mean value for all datasets; and that of categorical features takes the mean value for the Census Income and German Credit dataset, and takes the mode for the Cretio dataset. |