Path Choice Matters for Clear Attributions in Path Methods

Authors: Borui Zhang, Wenzhao Zheng, Jie Zhou, Jiwen Lu

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we conduct qualitative and quantitative experiments to demonstrate the superiority of our proposed SAMP method.
Researcher Affiliation Academia Borui Zhang , Wenzhao Zheng , Jie Zhou , Jiwen Lu Department of Automation, Tsinghua University, China {zhang-br21, zhengwz18}@mails.tsinghua.edu.cn; {jzhou, lujiwen}@tsinghua.edu.cn
Pseudocode Yes Algorithm 1: The SAMP++ algorithm.
Open Source Code Yes Corresponding author. 1Code: https://github.com/zbr17/SAMP
Open Datasets Yes We evaluate SAMP on the widely used MNIST (Deng, 2012), CIFAR-10 (Krizhevsky et al., 2009), and Image Net (Deng et al., 2009).
Dataset Splits No The paper mentions training models and using test sets, but does not explicitly provide details about specific training/validation/test dataset splits (e.g., percentages or counts) or reference predefined splits with citations for reproducibility beyond stating the datasets used.
Hardware Specification Yes We perform all experiments with Py Torch on one NVIDIA 3090 card.
Software Dependencies No The paper mentions 'Py Torch torchvision package' and 'Adam W optimizer' but does not specify exact version numbers for these software dependencies to ensure reproducibility.
Experiment Setup Yes If without special specifications, we fix the step size s in SAMP as 224 16 for Image Net and 10 for other datasets, the ratio of the infinitesimal upper bound η to x 1 as 0.1, and the momentum coefficient λ as 0.5.