Get input required for a given model output

Web: https://www.reddit.com/r/deeplearning/comments/vhcbn0/get_input_required_for_a_given_model_output/

Hello Folks! I'm gathering information on how to obtain the scope of inputs (it can be more than one) required for a given output. Let's suppose I'm using a vanilla 1 hidden layer fully connected network with non linear activation function/

I've come across a few options like, numerically solving the inverse equation (given its non linearity, not sure how one would solve analytically, but we can analytically end up with multiple equations from relu's..), using backpropagation with a defined …

deeplearning model