Degradation of validation accuracy

[Sheethal00]

Hi,

I have pruned 5% of filters from the facenet model having least l1 norm and calculated the validation accuracy for pruned model on lwf dataset, i am getting 6%

In order to resolve this i assigned zero to the pruned filters and did not prune those filters, with this change i am getting 96% vaidation accuracy.

When i compared the weights of these two models they are matching but the validation accuracy is different

Can u please help me with this

Regards Sheethal