Automated Scoring on NGSS-aligned modeling Item 4 (test)
82.1AccuracyViT + LoRA
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ViT + LoRABackbone=vit_base_patch16_224, Fine-tuning strategy=LoRA, Confidence-aware scoring mechanism=None2026.06 | 82.1 | 78.8 | 72.1 | 70.2 | 70.8 | |
| CA-UniformBackbone=vit_base_patch16_224, Fine-tuning strategy=LoRA, Confidence-aware scoring mechanism=Uniform2026.06 | 82.1 | 81.3 | 70.8 | 67.9 | 68.7 | |
| CA-SelectiveBackbone=vit_base_patch16_224, Fine-tuning strategy=LoRA, Confidence-aware scoring mechanism=Selective2026.06 | 82.1 | 81.3 | 70.8 | 67.9 | 68.7 | |
| ViT (Frozen)Backbone=vit_base_patch16_224, Fine-tuning strategy=Frozen, Confidence-aware scoring mechanism=None2026.06 | 24.4 | 4 | 41.1 | 34.7 | 15.2 |