“# SUPERVISED-CONTRASTIVE-LEARNING-FOR-PRE-TRAINED-LANGUAGE-MODEL-FINE-TUNING”
in this code, I’ve implemented sentiment analysis task with sst-2 dataset.
the below results are for 100 training samples:
cross entropy loss:
cross entropy + contrastive loss:
cross entropy heatmap on test dataset:
Accuracy on test dataset: 90.13
cross entropy + contrastive loss heatmap:
Accuracy on test dataset: 92.20
paper:
https://arxiv.org/abs/2011.01403
