Whether it is difficult to train or debug？

[HalvesChen]

Thank you for your excellent work, I would like to ask if the training of the model is stable after using IBQ? If you can, please provide some experimental considerations

[ShiFengyuan1999]

Hi @HalvesChen. Yes, the training of IBQ model is very stable. There are no sudden loss spikes in all our experiments. Actually, it is easy to train IBQ and get good results without special training tricks. Of course, you can also apply some training tricks, such as LeCAM regularization, to further improve the stability of the model training.

[HalvesChen]

Thanks~

[HalvesChen]

Hi,
I notice that your method requires much more epoch(330) than other methods. Have you compared the reconstruction performance of the common setting (epoch=50)?