Inverse xgboost for estimation

Hello all, Consider we are using xgboost regressor. We have some inputs and outputs and we train our system with our dataset. Also, the accuracy of the xgboost learner is acceptable.
Here is my question: now given an output, are there any ways to find the inputs?
I mean, typically, after we trained our xgboost learner, we give some new inputs to it and we are interested in finding the relevant outputs. But, can we do this process inversely? Any suggestions?