Array mutating problem when calculating inverse of (I + M)

[MrstupidJ]

RobustNeuralNetworks.jl/src/ParameterTypes/passive_ren.jl

Line 115 in 67f9f81

else

There is an array mutating in $\rho \neq 0$ case. This will not affect constructing a model with the corresponding robust property, however, this will affect learning/training of the model via Zygote.

Shall be fixed ASAP.

[MrstupidJ]

The problem is likely introduced when we calculate D22 (this line) .

Not sure about the fix yet, but we could use general paramterisation for now.

[nic-barbara]

@MrstupidJ we use the following function to calculate D22:

_D22_pass(M, ρ) = ((I+M) \ I) / ρ

As far as I'm aware that shouldn't be mutating any arrays.

Could you please provide a minimal example reproducing the error to help with debugging?

[MrstupidJ]

Applied the same fix as here.