Ensemble Predictor

Ensemble Configuration

CNN-Small 0.20

Accuracy: 92.3% Speed: Fast Params: 1.2M Specialty: Basic

CNN-CBAM 0.25

Accuracy: 94.8% Speed: Medium Params: 2.5M Specialty: Attention

ResNet-18 0.30

Accuracy: 95.6% Speed: Medium Params: 11M Specialty: Deep

ViT-Tiny 0.15

Accuracy: 93.7% Speed: Slow Params: 5.7M Specialty: Global

EfficientNet 0.25

Accuracy: 96.2% Speed: Medium Params: 5.3M Specialty: Efficient

MobileNet 0.10

Accuracy: 91.5% Speed: Very Fast Params: 3.2M Specialty: Mobile

Voting Strategy

Uncertainty Method

Confidence Threshold 70%

Wafer Scratch

85.3%

Particle Defect

8.2%

Ring Pattern

4.1%

Edge Location

2.4%

0.23

Uncertainty Score

Low Uncertainty - High Confidence

ResNet-18 28.5%

EfficientNet 24.2%

CNN-CBAM 23.8%

CNN-Small 18.5%

ViT-Tiny 5.0%