I have read this paper about the effects of quantum noise on QAOA.
In the conclusion, it says: QAOA is a noise-tolerant algorithm, quantum noise does not change the QAOA quantum circuit parameter optimization direction, for both noisy and ideal QAOA cost functions, the optimized QAOA parameters being nearly the same.
I use qiskit to perform some experiments. Basically, I add a noise model to the simulator during the parameter training process. I found that, the optimized parameters are different when I use qasm_simulator and COBYLA.
Is the difference of the optimized parameters caused by the sampling noise for qasm_simulator? Will the noise model impact the value of the optimized parameters theoretically?