Matrix inverse approximation with keras dense model

Question

I am training a neural network to calculate the inverse of a 3x3 matrix. I am using a Keras dense model with 1 layer and 9 neurons. The activation function on the first laye