Experiment Log: Comparing current findings with orignal paper

Date

2026-01-20

Hypothesis

The current architecture should produce the same results as the original paper.

Experimental Setup

• Model/Architecture: LeNet-1
• Dataset: MNIST
• Preprocessing: normalization, resizing (28x28)
• Hyperparameters:
  • Learning rate: 0.01
    
  • Batch size: 32
    
  • Epochs: 20
    
  • Loss Function: Cross Entropy Loss
    
  • Others:

Procedure

• Using Pytorch to get the MNIST data.
• Setting:
  • The input image size to 28x28.
  • The CrossEntropyLoss() as the Loss function.
  • The optimizer: optimizer = optim.SGD(model.parameters(),lr=0.01,momentum=0.9)
• Training on 20 epochs.

Results

• Training Loss: 0.0397
• Training Accuracy Rate: 98.80%
• Test Accuracy: 0.9868
• Test Error Rate: 1.32%
• Other Metrics:

Visualizations

Observations

• I notice that I got a lower error rate for the test dataset, although the same architecture and procedures are used. However this is expected range for the network due to its small size.

Conclusions

• The results are expected.

Next Steps / Ideas

• Rerun the training using the max pooling insted of the average.
• Try with ReLU insted of Tanh.
• Add padding.

---