In fields like machine learning, where the validity of your solution is measured on the basis of the accuracy of your model, it is important to have a simulation i.e. write and run code to implement the machine learning algorithm you're working on. However, if I am just proposing a new framework, is it necessary for me to run a simulation for that? What else can I do to evaluate the validity of my solution?
Yes, it seems in your case simulation is necessary. In many cases, not probably for the situation involving machine learning, simulation results must be backed by mathematical analysis and proofs.
If you have proposed a framework or an idea, how would anyone know if it is workable? Science requires hard proofs backed by experiments. So, Yes it is necessary for you to run a simulation.