Visualizing the Impact of Feature Attribution Baselines

By Pascal Sturmfels, Scott M. Lundberg, Su-In Lee. 2020

314 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5ca973f48d42ab33f798bee42f2af74f9cdd8d26

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Interpreting Super-Resolution Networks with Local Attribution Maps (2020, methodology): “However, the gradient method suffers from the “saturation” issue [43, 42], which could lead to the problem – the graF”
Interpreting Graph Neural Networks for NLP With Differentiable Edge Masking (2020, background): “This could harm the performance of integrated gradients (Sturmfels et al., 2020); however, as we have constructed our synthetic task such that the number of leaves and thus the degree of the centroid changes, a GNN which achieves a perfect score for this task must be robust with respect to changing…”
Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications (2021, methodology): “In Section IV, we experiment with a combination of integrated gradients and SmoothGrad [176], similar to expected gradients (see [181]), where relevance scores obtained from integrated gradients are averaged over several integration paths that are drawn from some random distribution.”
Algorithms to estimate Shapley value feature attributions (2022, background): “For example, in images, removing features with a baseline cannot give credit to pixels that match the baseline [26, 29, 44]; for a mean baseline, this means that regions of images that resemble the mean will be biased towards lower importance.”
How Do Decisions Emerge across Layers in Neural Models? Interpretation with Differentiable Masking (2020, background): “…can prevent users from trusting model predictions (Kim, 2015; Ribeiro et al., 2016), makes it hard to detect model or data deﬁciencies (Gururangan et al., 2018; Kaushik and Lipton, 2018) or verify that a model is fair and does not exhibit harmful biases (Sun et al., 2019; Holstein et al., 2019).”