Visualization of Nonlinear Classification Models in Neuroimaging - Signed Sensitivity Maps
By Peter Mondrup Rasmussen, Tanya Schmah, Kristoffer Hougaard Madsen, Torben Ellegaard Lund, G. Yourganov, Stephen C. Strother, Lars Kai Hansen. 2012
22 citations recorded by the source corpus.
Classification models are becoming increasing popular tools in the analysis of neuroimaging data sets. Besides obtaining good prediction accuracy, a competing goal is to interpret how the classifier works. From a neuroscientific perspective, we are interested in the brain pattern reflecting the underlying neural encoding of an experiment defining multiple brain states. In this relation there is a great desire for the researcher to generate brain maps, that highlight brain locations of importance to the classifiers decisions. Based on sensitivity analysis, we develop further procedures for model visualization. Specifically we focus on the generation of summary maps of a nonlinear classifier, that reveal how the classifier works in different parts of the input domain. Each of the maps includes sign information, unlike earlier related methods. The sign information allows the researcher to assess in which direction the individual locations influence the classification. We illustrate the visualization procedure on a real data from a simple functional magnetic resonance imaging experiment.
Subfields: Artificial Intelligence, Cognitive Neuroscience
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
- Wojciech Samek
- K. Müller
- Grgoire Montavon
- Sebastian Lapuschkin
- Alexander Binder
- Christopher J. Anders
- Longo Luca
- Giulia Vilone
In Cognitive Neuroscience
- Grgoire Montavon
- Sebastian Lapuschkin
- Wojciech Samek
- K. Müller
- Alexander Binder
- Anna Choromanska
- Benjamin Blankertz
- Cuntai Guan
Citations in the corpus
- Evaluating the Visualization of What a Deep Neural Network Has Learned (2015, background): “Further previous work has focused on understanding nonlinear learning methods such as DNNs or kernel methods [14], [25], [26] essentially by sensitivity analysis in the sense of scores based on partial derivatives at the given sample.”
- VisualBackProp: efficient visualization of CNNs (2018, background): “Finally, sensitivity-based methods [7, 8, 9]) aim to understand how the classifier works in different parts of the input domain by computing scores based on partial derivatives at the given sample.”