Multi instance NN modeling and training, dataset only contains each instance inputs and global output (unknowed hidden individual outputs)

Question

Is it possible to train multi-instance neural network (all instances with identical parameters), in which each instance recives its own vector of inputs, and generates a hid