Add FlatRule and WSquareRule

Author: adrhill

src/ExplainableAI.jl

@@ -42,7 +42,7 @@ export LRP
 # LRP rules
 export AbstractLRPRule
 export LRP_CONFIG
-export ZeroRule, EpsilonRule, GammaRule, ZBoxRule, PassRule
+export ZeroRule, EpsilonRule, GammaRule, WSquareRule, FlatRule, ZBoxRule, PassRule
 export modify_input, modify_denominator
 export modify_param!, modify_layer!
 export check_model

src/lrp_rules.jl

@@ -156,6 +156,29 @@ function modify_param!(r::GammaRule, param::AbstractArray{T}) where {T}
     return nothing
 end
 
+"""
+    WSquareRule()
+
+LRP-``W^2`` rule. Commonly used on the first layer when values are unbounded.
+
+# References
+[1]: G. Montavon et al., Explaining nonlinear classification decisions with deep Taylor decomposition
+"""
+struct WSquareRule <: AbstractLRPRule end
+modify_param!(::WSquareRule, p) = p .^= 2
+modify_input(::WSquareRule, input) = ones_like(input)
+
+"""
+    FlatRule()
+
+LRP-Flat rule. Similar to the [`WSquareRule`](@ref), but with all parameters set to one.
+
+# References
+[1]: S. Lapuschkin et al., Unmasking Clever Hans predictors and assessing what machines really learn
+"""
+struct FlatRule <: AbstractLRPRule end
+modify_param!(::FlatRule, p) = fill!(p, 0)
+modify_input(::FlatRule, input) = ones_like(input)
 
 """
     PassRule()
@@ -238,7 +261,7 @@ for R in (ZeroRule, EpsilonRule)
 end
 
 # Fast implementation for Dense layer using Tullio.jl's einsum notation:
-for R in (ZeroRule, EpsilonRule, GammaRule)
+for R in (ZeroRule, EpsilonRule, GammaRule, WSquareRule, FlatRule)
     @eval function lrp!(Rₖ, rule::$R, layer::Dense, aₖ, Rₖ₊₁)
         reset! = get_layer_resetter(rule, layer)
         modify_layer!(rule, layer)

src/utils.jl

@@ -52,6 +52,29 @@ CartesianIndex(5, 3)
 """
 drop_batch_index(C::CartesianIndex) = CartesianIndex(C.I[1:(end - 1)])
 
+"""
+    ones_like(x)
+
+Returns array of ones of same shape and type as `x`.
+
+## Example
+```julia-repl
+julia> x = rand(Float16, 2, 4, 1)
+2×4×1 Array{Float16, 3}:
+[:, :, 1] =
+ 0.2148  0.9053   0.751    0.358
+ 0.38    0.09033  0.04053  0.6543
+
+julia> ones_like(x)
+2×4×1 Array{Float16, 3}:
+[:, :, 1] =
+ 1.0  1.0  1.0  1.0
+ 1.0  1.0  1.0  1.0
+```
+"""
+ones_like(x::AbstractArray) = ones(eltype(x), size(x))
+ones_like(x::Number) = oneunit(x)
+
 # Utils for printing model check summary using PrettyTable.jl
 _print_name(layer) = "$layer"
 _print_name(layer::Parallel) = "Parallel(...)"