As my final year project I am working on a Quantum GAN for drug discovery, however the training process is quite slow on a simulator, takes about 1.5 hours to complete one epoch using 6 qubits. I am quite short on time and resources to complete it.
I wanted to know if it is better to train my model first on the simulator or just train the model directly on a real quantum computer.
I am using IBM’s brisbane device
Also, since the 10 minutes provided when using IBM’s device is a very short amount of time, are there alternatives?
Hi @manik1080 !
Welcome to the forum.
I don’t have a definite answer to your question, it is tricky, but I can provide a few suggestions.
- 1.5 hours seems like a long time for only six qubits, I would consider going through your code using a Python profiler (e.g. SnakeViz) to see where it is taking so long.
- I believe that as a sanity check it is better to train your model first on the simulator and once you are sure about it, run it on a real QC. I mean debugging and fine tuning your model as much as possible before running it.
Finally, if your ultimate goal is to run your model on a real QC and you have time constraints, you can consider running shorter experiments that allow you to have an idea of the impact of noise on your model’s performance.
Good luck with this. Let us know if you have more questions.