Visualizing Deep Networks by Optimizing with Integrated Gradients

Authors: Zhongang Qi, Saeed Khorram, Li Fuxin11890-11898

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on several benchmark datasets show that the heatmaps produced by our approach are more correlated with the decision of the underlying deep network, in comparison with other state-of-the-art approaches.
Researcher Affiliation Collaboration 1School of Electrical Engineering and Computer Science, Oregon State University 2Applied Research Center, PCG, Tencent zhongangqi@tencent.com, {khorrams, lif}@oregonstate.edu
Pseudocode Yes Algorithm 1: I-GOS
Open Source Code No The paper does not provide an explicit statement or a link to the open-source code for the methodology described.
Open Datasets Yes We utilize the pretrained VGG19 (Simonyan and Zisserman 2015) and Resnet50 (He et al. 2016) networks from the Py Torch model zoo to test 5, 000 randomly selected images from the validation set of Image Net (Russakovsky et al. 2015).
Dataset Splits Yes We utilize the pretrained VGG19 (Simonyan and Zisserman 2015) and Resnet50 (He et al. 2016) networks from the Py Torch model zoo to test 5, 000 randomly selected images from the validation set of Image Net (Russakovsky et al. 2015).
Hardware Specification Yes For each approach, we only use one Nvidia 1080Ti GPU.
Software Dependencies No The paper mentions 'Py Torch model zoo' but does not specify version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes In Eq. (8), β = 0.0001. λ1 and λ2 in Eq. (9) were fixed across all experiments under the same heatmap resolution. We downloaded and ran the code for most baselines, except for (Sundararajan, Taly, and Yan 2017) which we implemented. All baselines were tuned to best performances. For RISE, we followed (Petsiuk, Das, and Saenko 2018) to generate 4, 000 7 7 random samples for VGG, and 8, 000 7 7 random samples for Res Net. For IGOS, the maximal iteration is 15; for Mask, the maximal iteration is 500.