Fine-Tuning Pre-trained Language Model with Weak Supervision: A Contrastive-Regularized Self-Training Approach

Benchmark Model Rank Results
sentiment-analysis-on-imdbCOSINE#34Accuracy: 90.54
text-classification-on-yelp-2COSINE#5Accuracy: 95.97%
word-sense-disambiguation-on-words-in-contextCOSINE + Transductive Learning#1Accuracy: 85.3