batch size affect

[VietHoang1512]

Hi, I'm trying to experiment with CIFAR-10 with the default hyper-params, and it seems to yield a better score when using smaller batch size (e.g. 72% with batch size 256 yet 78% with batch size 128). Anyone in the same situation, here?

[VietHoang1512]

STL10 with default hyper-params experients yields the similar results (90.8203% with batch size 256 yet 91.7969% with batch size 128)

[Arandinglv]

Maybe overfitting? Small batch-size means fewer negative samples...