OpflowQNN optimization with forward and backward pass

I am learning Quantum Machine Learning with qiskit OpflowQNN. I understand how to set up the problem with parametrized circuit and cost operator, following the example provided by qiskit github.com/Qiskit/qiskit-machine-learning/blob/main/docs/tutorials/01_neural_networks.ipynb

I wonder how do I optimize the cost operator/function? Should I write a loop to perform forward qnn1.forward(input1, weights1) and backward qnn1.backward(input1, weights1) operations till the cost operator/function converge? Any examples?