Normalization & Regularization

Code for the Experiments

google-colab link: 06_NormalizationRegularization_Modularized_FINAL

github link: 06_NormalizationRegularization_Modularized_FINAL.ipynb

Dataset: MNIST Digits
Model: CNN - 9960 params
Epochs: 40

Regularization Values

l1_lambda = 5e-4
l2_lambda = 5e-3

L1 Regularization

L2 Regularization

Results

1. No L1 No L2

Highest Accuracy: 99.47

Metrics

Misclassifications

2. L1

Highest Accuracy: 99.44

Metrics

Misclassifications

3. L2

Highest Accuracy: 99.52

Metrics

Misclassifications

4. L1 & L2

Highest Accuracy: 98.76

Metrics

Misclassifications

Comparison

Validation Accuracy

Validation Loss