Table 6 Ablation study results showing the effect of removing individual components from the proposed model on validation accuracy and F1 Score.

From: Resolving passage ambiguity in machine reading comprehension using lightweight transformer architectures

Architecture variant

Validation accuracy (%)

F1 Score (%)

Proposed full architecture

92.70

91.42

Without QW + EA

89.50

87.65

Without [CLS]

90.10

88.72

Without FC Layer

88.30

86.40

Without softmax

87.90

85.90

Original BERT

91.00

90.23

Back to article page