ExplainableAI.jl

This package implements interpretability methods for black box models, with a focus on local explanations and attribution maps in input space. It is similar to Captum and Zennit for PyTorch and iNNvestigate for Keras models. Most of the implemented methods only require the model to be differentiable with Zygote. Layerwise Relevance Propagation (LRP) is implemented for use with Flux.jl models.