LeNet-5 Training on MNIST Dataset with Enhanced Implementation
This resource implements the LeNet-5 architecture for the MNIST dataset, adapting the original network structure by modifying input dimensions to 28×28 pixels. The implementation draws inspiration from UFLDL tutorials and R. B. Palm's CNN codebase. Key modifications include full connectivity between C3 and S4 feature maps, achieving 99.1% accuracy through optimized training procedures with data augmentation and regularization techniques.