Hi, I recently conducted a few-shot fine-tuning on my private ECG classification training dataset and was surprised by the performance on the test dataset. This high performance occurred after just two epochs, regardless of whether I used an SSL pretrained model or not. My question is: does the model utilize the pretrained weights in both scenarios? Also, are some of the weights frozen during fine-tuning, contributing to the rapid fine-tuning?
Hi, I recently conducted a few-shot fine-tuning on my private ECG classification training dataset and was surprised by the performance on the test dataset. This high performance occurred after just two epochs, regardless of whether I used an SSL pretrained model or not. My question is: does the model utilize the pretrained weights in both scenarios? Also, are some of the weights frozen during fine-tuning, contributing to the rapid fine-tuning?