Skip to Main Content
Table 4: 
Classification accuracy with reduced-size encoders.
B → E A → K
5 layers8 layers10 layers12 layers (full)5 layers8 layers10 layers12 layers (full)
BERT 70.9 75.9 80.6 78.8 71.2 74.9 81.2 78.8 
Fine-tuned BERT 74.6 76.5 84.2 84.2 74.0 76.3 80.8 81.9 
PERL (Ours) 81.1 83.2 88.2 87.0 77.7 80.2 84.7 84.2 
B → E A → K
5 layers8 layers10 layers12 layers (full)5 layers8 layers10 layers12 layers (full)
BERT 70.9 75.9 80.6 78.8 71.2 74.9 81.2 78.8 
Fine-tuned BERT 74.6 76.5 84.2 84.2 74.0 76.3 80.8 81.9 
PERL (Ours) 81.1 83.2 88.2 87.0 77.7 80.2 84.7 84.2 
Close Modal

or Create an Account

Close Modal
Close Modal