Saliency-Based explanations via generative modelling
A framework for explaining deep neural networks for neuroimaging via generative modelling based on large-scale data.
In this project, we propose a framework for explaining deep neural networks for neuroimaging via generative modelling based on large-scale data. The idea is to ground the explanations in the data distribution, and to use the generative model to generate synthetic data that is as similar as possible to the real data.
The project is under construction and will be published soon :)