InstaSHAP: Interpretable Additive Models Explain Shapley Values Instantly

Authors: James Enouen, Yan Liu

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 6, the paper presents "TABULAR EXPERIMENTS" with subsections "SYNTHETIC EXPERIMENTS", "SYNERGY IN BIKE SHARING", and "REDUNDANCY IN TREE COVER". In Section 7, it presents "HIGHER DIMENSIONAL EXPERIMENTS" with the text: "Fine-tuning a resnet-50 model on the dataset, we are able to achieve a fine-grained accuracy of 65.0% and a coarse-grained accuracy of 81.8%." These sections describe empirical studies, dataset evaluation, and performance metrics.
Researcher Affiliation Academia Both authors, James Enouen and Yan Liu, are affiliated with the "Department of Computer Science, University of Southern California, Los Angeles, CA" and their email addresses are "EMAIL" and "EMAIL", respectively. These affiliations are all academic.
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks. It primarily uses mathematical equations and descriptive text to explain methodologies.
Open Source Code No The paper does not contain any explicit statement about releasing source code, nor does it provide a link to a code repository in the main text or appendices.
Open Datasets Yes In Section 7, the paper states: "We additionally use our methods to explore a bird classification task on natural images. We run experiments on the CUB dataset using a resnet CNN architecture... CUB dataset consisting of 200 different species of birds and containing over 6000 labeled images (Wah et al., 2011)." The CUB dataset is a well-known public dataset, and the paper provides a proper citation for it.
Dataset Splits No The paper mentions evaluating models and reporting accuracies (e.g., "Fine-tuning a resnet-50 model on the dataset, we are able to achieve a fine-grained accuracy of 65.0% and a coarse-grained accuracy of 81.8%") and validation accuracy (e.g., "Training an MLP on this dataset is able to achieve validation accuracy of 80.4%"), implying the use of dataset splits. However, it does not provide specific percentages or sample counts for these splits, nor does it cite predefined standard splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments, such as GPU models, CPU types, or memory specifications. It only mentions using a "resnet CNN architecture" generally.
Software Dependencies No The paper mentions using a "resnet CNN architecture" but does not specify any software libraries (e.g., PyTorch, TensorFlow) or their version numbers that would be necessary to reproduce the experiments.
Experiment Setup Yes In Appendix D.4, for the computer vision experiments, the paper states: "Both models are initialized with mostly pretrained weights and fine-tuned for 300 epochs on the CUB dataset." This provides a specific hyperparameter (number of epochs) used in the experimental setup.