Full-Gradient Representation for Neural Network Visualization

Authors: Suraj Srinivas, François Fleuret

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally evaluate the usefulness of Full Grad in explaining model behaviour with two quantitative tests: pixel perturbation and remove-and-retrain.
Researcher Affiliation Academia Suraj Srinivas Idiap Research Institute & EPFL suraj.srinivas@idiap.ch François Fleuret Idiap Research Institute & EPFL francois.fleuret@idiap.ch
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described in this paper.
Open Datasets Yes We use our pixel perturbation test to evaluate full-gradient saliency maps on the Imagenet 2012 validation dataset, using a VGG-16 model with batch normalization. We use ROAR to evaluate full-gradient saliency maps on the CIFAR100 dataset, using a 9-layer VGG model.
Dataset Splits Yes We use our pixel perturbation test to evaluate full-gradient saliency maps on the Imagenet 2012 validation dataset, using a VGG-16 model with batch normalization.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup No The paper mentions models used (VGG-16, 9-layer VGG) and some post-processing parameters (e.g., bilinear Upsample, rescale, abs) for saliency map generation, but it does not provide specific hyperparameters (e.g., learning rate, batch size, number of epochs, optimizer settings) or system-level training settings for the neural networks.