Sanity Checks for Saliency Maps

Authors: Julius Adebayo, Justin Gilmer, Michael Muelly, Ian Goodfellow, Moritz Hardt, Been Kim

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments we show that some existing saliency methods are independent both of the model and of the data generating process.
Researcher Affiliation Collaboration Julius Adebayo , Justin Gilmer , Michael Muelly , Ian Goodfellow , Moritz Hardt , Been Kim juliusad@mit.edu, {gilmer,muelly,goodfellow,mrtz,beenkim}@google.com Google Brain University of California Berkeley
Pseudocode No The paper describes methods in prose and through mathematical formulations but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes All code to replicate our findings will be available here: https://goo.gl/h Bmh Dt
Open Datasets Yes Inception v3 model trained on Image Net.; CNN Fashion MNIST MLPMNIST Inception v3 Image Net; MNIST test set for a CNN.
Dataset Splits No The paper mentions training models on datasets like ImageNet, MNIST, and Fashion MNIST, and refers to a “test set,” but does not provide specific training, validation, or test split percentages, sample counts, or explicit splitting methodologies for reproducibility.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or exact cloud instance types) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies, such as library names with version numbers (e.g., TensorFlow 2.x, PyTorch 1.x), required to replicate the experiments.
Experiment Setup No The paper describes the general training process for models (e.g., training to >95% accuracy) but does not provide specific experimental setup details such as hyperparameters (learning rate, batch size, number of epochs, optimizer settings) or detailed training configurations.