I am using a dataset for classification consists of nearly ~20k images.

I want to train them faster as a result I used EfficientNet. But the performance keeps dropping while I am using better versions of the architecture consists of a greater number of parameters. cross-validation acc was 0.836 for version B0 when 0.732 for version B5.

Can anyone explain to me why increasing of parameter making the result fall?

More Tashin Ahmed's questions See All
Similar questions and discussions