hard20 pts · first answer countsnormalizationfine-tuning

Fine-tuning a CNN with batch size 2

You fine-tune a pre-trained CNN with BatchNorm using batch size 2 because of memory limits. Validation accuracy is erratic and worse than before fine-tuning. What is the most likely fix?

A.Raise the dropout rate to 0.9B.Use a much larger learning rateC.Remove the final classification layerD.Freeze BatchNorm statistics or switch to GroupNorm

Discussion 0

Join free to post in the discussion.

Fine-tuning a CNN with batch size 2

Discussion 0

Related questions