Controllably Sparse Perturbations of Robust Classifiers for Explaining Predictions and Probing Learned Concepts

dc.contributor.authorRoberts, Jayen_US
dc.contributor.authorTsiligkaridis, Theodorosen_US
dc.contributor.editorArchambault, Daniel and Nabney, Ian and Peltonen, Jaakkoen_US
dc.date.accessioned2021-06-12T11:28:25Z
dc.date.available2021-06-12T11:28:25Z
dc.date.issued2021
dc.description.abstractExplaining the predictions of a deep neural network (DNN) in image classification is an active area of research. Many methods focus on localizing pixels, or groups of pixels, which maximize a relevance metric for the prediction. Others aim at creating local "proxy" explainers which aim to account for an individual prediction of a model. We aim to explore "why" a model made a prediction by perturbing inputs to robust classifiers and interpreting the semantically meaningful results. For such an explanation to be useful for humans it is desirable for it to be sparse; however, generating sparse perturbations can computationally expensive and infeasible on high resolution data. Here we introduce controllably sparse explanations that can be efficiently generated on higher resolution data to provide improved counter-factual explanations. Further we use these controllably sparse explanations to probe what the robust classifier has learned. These explanations could provide insight for model developers as well as assist in detecting dataset bias.en_US
dc.description.sectionheadersPapers
dc.description.seriesinformationMachine Learning Methods in Visualisation for Big Data
dc.identifier.doi10.2312/mlvis.20211072
dc.identifier.isbn978-3-03868-146-5
dc.identifier.pages1-5
dc.identifier.urihttps://doi.org/10.2312/mlvis.20211072
dc.identifier.urihttps://diglib.eg.org:443/handle/10.2312/mlvis20211072
dc.publisherThe Eurographics Associationen_US
dc.subjectComputing methodologies
dc.subjectMachine learning
dc.subjectArtificial intelligence
dc.titleControllably Sparse Perturbations of Robust Classifiers for Explaining Predictions and Probing Learned Conceptsen_US
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
001-005.pdf
Size:
658.38 KB
Format:
Adobe Portable Document Format